I've been running a lot of sims recently, and I thought I'd share a few of my conclusions.

Talents

Dark Arts should probably be marked as ‘required’ - shadow bite contributes 20% of your dps, so this is a 1% dps boost per point, making it stronger than any of the destro talents.

Master summoner is better than you might think. Even if all your demon swaps are instant, because of soulburn/demonic rebirth, the mana saving is huge, which is likely to mean you avoid life-tapping, and thus gain dps. It’s not correctly implemented on SimulationCraft (I’ve submitted a bug report, and it will be fixed in the next release), but I conservatively estimate that with demon-swapping, master summoner gives a gain of 0.35% dps/point.

Glyphs

Glyph of corruption is easily better than glyph of incinerate. With my gear (which is not great) corruption is worth 130 dps, and incinerate is worth 50 dps. Corruption is going to be even more of an improvement in movement phases, because it gives you an additional instant-cast spell, and it’s also more fun. Of course this means that shadow bolt should take priority above incinerate and soul fire when shadow trance procs.

Glyph of felguard is amazingly poor. It’s only a 5% damage increase (compare 25% for lash of pain and 20% for imp) and it only applies to an attack which at best contributes a third of your pet’s damage. It doesn’t scale with multiple opponents, since the damage from legion strike is divided between them. I make it a gain of only 64 dps, which means it’s inferior to corruption even if you’re using your felguard for the whole fight. (Of course, corruption would be useless if you were solely using inferno, but it will be very strong with multi-dotting.)

The value of glyph of life tap is questionable. You gain so much mana from mana feed, that you will probably need life tap very rarely, in which case it would be better to take both soul link and soulstone. However, if you find you are using life tap a lot, then you should certainly glyph it.

Stat priorities

I suspect that the suggestion that haste is better than mastery at lower gear levels is now out of date (with the buff to the doomguard in patch 4.1). With my gear, which isn’t great, the sims are giving me 1.46 dps per point of mastery and 1.20 dps per point of haste.

Demon swapping

The felguard’s demon soul effect is approximately 4 times as good as the felhunter’s demon soul effect. Various people have suggested that this means it would be superior to start off with the felguard out, hit demon soul and felstorm, and then switch to the felhunter, but I haven’t seen anyone give any numbers.

My sims give me 24200 dps with a pure felhunter rotation, as opposed to 24540 using an initial felguard (over a seven and half minute fight), so the dps gain is significant but not huge.

A further dps gain is available by summoning the felguard every 2 minutes, hitting demon soul and felstorm, and then recalling the felhunter. Optimal output is obtained by hard summoning the felguard when demon soul is 5s off CD and using soulburn for the felhunter once felstorm is complete. This gives me 24920 dps.

Whether you think a total gain of 3% dps is worth the effort is a matter for you.

Kardacz suggested on 4th March that swapping demons in combat required you to recast soul link. I’ve just tested this, and it appears that it is no longer the case. Provided that you cast soul link on each demon in turn before combat, you shouldn’t need to recast.

Cooldown synergies

Trinkets like soul casket which have an on-use ability with a 2 minute cooldown and a 20 second duration work out very well, because this exactly matches demon soul, which means you can just macro them together and always get the synergy. (There is, in any case, something very satisfactory about matching a soul casket with a demon soul.)

BoD has a 1 minute duration, so it naturally synchronises with demon soul, meaning that 50% of your BoDs benefit from the buff. Letting them get out of sync will result a fairly serious dps loss, so if you delay demon soul to wait for metamorphosis, you must also delay BoD.

Fortunately, you scarcely ever need to delay demon soul, because metamorphosis has an average cooldown of 106s. However, since metamorphosis is much stronger buff than the felhunter demon soul, the benefit of the synergy gained by delaying metamorphosis is more than counteracted by the loss of metamorphosis uptime. So in a pure felhunter fight, optimal dps will be obtained by using each ability on CD.

If you’re using the felguard demon soul, then the sims show that waiting either metamorphosis or demon soul (plus BoD) for up to 20s will result in a slight gain in dps. (Of course since metamorphosis has a 36s duration and demon soul has a 20s duration, it is only necessary to wait metamorphosis until 16s before demon soul comes off CD.) Slight here means an increase of about 0.1% dps. Since using each ability on CD is much simpler to execute, I would suggest that aiming for the synergy is not worthwhile.

Of course there are plenty of fight specific situations where it would be worth delaying one or both CDs.

Your math is incorrect. You wouldn't lose the DPS attributed to corruption, immolate, BoD, doomgaurd, and if you timed it right you couldn't even lose out on the DPS attributed to HoG or incinerate. Basically you miss out on shadowbolt/soulfire, and your pet DPS (since your pet is stunned while you hard cast) which would be considerably lower than your overall dps.

I've simmed it thus:

Feel free to test with your own gear.

Your numbers don't look ridiculous as a first estimation. I'm losing 1.2s of casting time for the felguard start too, which on your analysis would be 65 dps loss, which means that the required gain from the initial felguard is 405 dps, which is hardly that far from 428. In any case, it's wrong to just multiply the time averaged dps by the time lost to calculate the damage lost, because that fails to take into account the spikiness of our damage output. That's the whole point of doing sims, rather than sketching calculations on the back of an envelope.

To be honest, I was surprised that swapping demons in mid-fight is a dps gain, but that is what the sims show, and a careful analysis of the logs makes me think that I am accurately simming what I'm aiming to sim, save that I'm not getting the mana gain from master summoner, which actually means the dps gain will be an underestimate.

Actually, I think SimulationCraft isn't taking that into account. Making exactly the sort of back-of-the-envelope calculation I just sneered at, I should be losing 13s uptime from a 6420 dps pet, which would be a loss of about 130 dps. If I then added back in the 85 dps I think I'm losing from the incorrect implementation of master summoner, my actual dps with the multi-swaps would be 24875, which is still better than 24540.

My mistake: the implementation on SimulationCraft is correct.

I saw these Simulationcraft Results recently, which use T12 heroics, and they imply that the stat priority should be haste>mastery>crit. This strikes me as incredibly strange since my understanding was that in heroic T11 gear mastery was best and I would have thought that trend would continue with even more powerful gear, but apparently not. Anyway, it might be good to change the main post to reflect this new stat priority information unless there is something I am missing.

Incidentally, there are a number of simple improvements which can be made to the priority list in Jadescorpion's link. Edit: The priority list used in the link has now been changed to substantially reflect these suggestions.

0. Edit - Eph has kindly pointed out that the volcanic potion should be used during metamorphosis, rather than during bloodlust. This gives a improvement of 74 dps.
1. We should try to summon the doomguard so as to take advantage of the initial pre-pot, bloodfury and the initial power torrent proc. Subject to that, we should wait for as long as possible, in order to allow necromantic focus time to stack. So we should summon either just before bloodfury expires (which will happen at 15s) or just before power torrent expires (which is random), whichever happens first. Once we know when to summon the doomguard, it should move to the top of priority list. Doing this gives a further improvement of 20 dps. Edit (11th August 2011) - When I originally posted this, the necromantic focus was simming as one of the two BiS trinkets. Currently, the darkmoon card: volcano is simming as superior, and it has accordingly been included in the BiS set on SimulationCraft. This means that stacking necromantic focus is no longer an issue, and we should aim to take advantage of (a) the initial pre-pot, (b) bloodfury, (c) the initial power torrent proc and (d) the initial DMC: V proc. At the moment I'm unable to get any dps improvement on time>10: that seems to be long enough that power torrent and DMC:V have always procced and short enough that nothing can fall off.
2. Hand of gul'dan is too low on the priority list, so we're losing uptime on the debuff. Also, immolate should be above BoD. Changing the priority to immolate, HoG, BoD gives a further improvement of 27 dps.
3. Shadow bolt should take priority over incinerate and soul fire when shadow trance procs. This gives a further improvement of 43 dps.
4. If we're going to be waiting for the moonwell chalice to come off CD before using metamorphosis, we might as well properly synchronise our cooldowns (remembering to delay BoD when demon soul is delayed). This gives a further improvement of 43 dps.
5. This isn't a 'simple' improvement, but we can still use demon swapping, as described in my earlier post. This gives a further improvement of 1317 dps.

Putting that all together gives this action list:
for a total improvement of 1514 dps. I'd be interested to know if anyone can improve on that.

(Incidentally, the sims in the link are run under version 420-3, which is not yet publically available, and which will fix a few bugs in 420-2, and thus gives slightly different dps numbers overall, but I don't think that will affect any of my suggested improvements.)

Edit - NB I have run these sims with all the buffs and debuffs selected in the Options pane. Different selections of buffs and debuffs will produce different results, and in particular, because of a (reported) bug in SimulationCraft, if you don't toggle on dark intent in the buff options, you will get incorrect results at step 1. All the above numbers are quoted on the basis of running 50,000 sims, but of course rerunning the sims will give results which will differ by a few dps either way.

Calculating the scale factors on that basis, haste still comes out much better than mastery, so the moral of the story is that you must calculate your own gear factors.

A further 107 dps can be gained by switching points from aura of foreboding(!) into master summoner.

He is suggesting Master Summoner (for the mana cost and cast time reductions) over those talent in conjunction with the modified action priority in the previous post that includes hard cast pet changes to allow for Felguard Demon Souls. If you are not taking advantage of that style of play then there will not be an increase. A point from Demonic Empowerment could be removed as well since it has very little raid use.

Eph's point is quite correct: the dps increase I quote is for the rotation I suggested in my previous post. For this purpose I'm concentrating entirely on maximising theoretical dps - I entirely agree that in practice the points may be better used somewhere else. However, it isn't correct to say that the benefit from Master Summoner only occurs if you hard cast a pet summon, since it also gives a 100% mana saving, which is well worth having, given that the summons cost 80% of base mana.

In my view, Aura of Foreboding is a poor raiding talent. It would be great if the effect only occurred when you actually wanted it, because, yes, there are lots of places it's advantageous to snare trash. However, you're going to be using HoG on CD, which means you'll be randomly snaring mobs every 12s, whether it's useful or not. I tank a lot, and random snares are second only random knockbacks in the list of annoying things dps do.

Aura of Foreboding is an unspeakably poorly designed talent for PvE considering that it's tied to a core rotational ability. In my opinion it should have been Soulburn: Hand of Gul'dan, possibly refunding the Shard if it actually stuns a target, but that's just wishlisting really.

In any case: Aura of Foreboding cannot really be considered a DPS increase, while the mana and/or time saving from Master Summoner makes it a much easier sell.

I'd be very surprised if this macro worked properly.

Firstly: all your targetting schenanigans are ruined by the fact that you're always instructing him to Legion Strike your current target, instead of his own target. I think this will cause problems if he's not actually in range of your target as well. I'm not sure why you'd want to macro in Legion Strike like this anway, surely it would be easier to have it auto-cast? The Felguard isn't chain-casting, so his Legion Strike is not vulnerable to the lag-induced cast delay experienced by the Imp if that's what you were trying to avoid. If you really want to have Legion Strike in your macro in this way you'll need a construction like the following (this also only acts if you've actually got a Felguard):

Secondly you can't have two "@" terms in a single conditional: you're only allowed at most one. Fortunately the /petattack command is one of those with a so-called "key unit" that allows you to check conditionals on the pet's target, then supply another one as the actual target of the attack, like so:

I hope this helps!