This is on par with what I suspected. It seems like taking an (ahem) extreme focus on either method simply takes away too much while giving too little. I can equate this to a simple metric in statistics called sensitivity versus specificity.

When you are diagnosing patients with a disease, you could have a 100% success rate of correctly diagnosing those with the disease by simply saying every single patient has the disease. The problem is, you are also incorrectly diagnosing those patients who do not have the disease. The goal, simply enough, is to find the best possible balance wherein your method correctly diagnoses those with the disease and correctly identifies those without the disease at the highest possible rate where both are balanced as best as possible. It will never be perfect, but the goal is to be as accurate on both parts as reasonably probable.

It applies in this situation: there are already variable outcomes when reaching certain stages of BG. Especially when reaching BG30, there is absolutely no way to determine without fail what will happen. In essence, the two scenarios already are playing themselves out regardless of what your intentions are, due to randomized factors such as proc rates. The goal being: to find that best possible balance between a pooling rotation and a rush through BG rotation. Oddly enough, in fights that are long enough it seems to me as though that balance is remarkably close to being found, regardless of what you intend to do.

I guess the TL;DR for this would be, I think running a proper rotation yields results that are measurably close to a theoretical max, regardless of what you plan to do with the BG transitions...AS LONG AS you understand the mechanics of them.

The developers did the following comment on Dancing Steels procchance:
"Dancing Steel: 1 PPM on melee attacks that land." (Source: Enchant Weapon - Dancing Steel - Spell - World of Warcraft)

In case this statement is correct, dancing steel shouldn't proc off poison applies or ticks.

If i don't lack knowledge, PPM usually favor slower weapons due to their higher individual proc chance on each hit and also on non-autoattack skills like sinister strike. In addition to this your actions per time are higher with combat than with assassination, so for now i got no real explanation on your observations. Especially with dual dancing steel with combat instead of only one.

My logs show the following uptimes this week with running dual dancing steel with two 2.6 weapons as combat.
Spirit Kings(Almost no phases of not hitting the bosses): 76.4%
Elegon(Some movement): 65.1%
My simcraft shows 50.29% uptime on my mainhands and 35.74% on my offhands dancing steel on a patchwerk fight.

The published reports for T14H on Simulationcraft Results show similar values to mine with combat(51%/34%) and do show smaller mainhand values for assassination(40%/37%), which should be correct according to my PPM statement above.

So this is a lot higher than your uptime for combat and a lot lower than your uptime for assassination.
You should probably try both setups on similar/equal circumstances like on the same boss or on the target dummy. I think you might have been just lucky on your high uptime as assassination(perhaps even the proc or WoL was/is buggy).

Keep in mind you get the 50% damage bonus from KS to your auto attacks made anyway, so the damage gain from shadow blades would be (shadow_blades_dmg - auto_attack_dmg)*0.5*number_of_hits.

My last log on the spirit kings encounter shows, that my shadow blades attack did 27k on average, while my normal hits did 13k on average while using two 2.6 speed weapons. I can get at most 6-8 auto attacks during a KS, resulting in a damage gain of 14k*7*0.5 = 49k damage total, whilst losing 3-3.5 seconds(thus 3-4 GCDs which are 0.8s with glyphed AR) of CP generators during AR.

In the second-worst case, the lack of those CP will lead to one less 5CP Eviscerate during deep insight, but one more during no insight. So one of my Eviscerates would lose 1 - 1/1.3 = of its damage. The logs show the exact amout of Evis' damage during specific infight phases, but according to my insight uptimes and average Evis damage it should be around 60k for no insight and 78000 for deep insight, losing only 18k dmg, compared to a 49k gain from KS.

In the worst case, you will lose a full Eviscerate over the fight and thus lose 60k+ damage total.

Seems like the difference is too minor to probably delay you're KS for some seconds to get it on par with SB.
This is of course just a random example, but unless somebody figures the exact DPS loss for CP losses, i can't tell better.

The Wowhead tooltip is bugged and you can reforge this massiv amount of mastery into haste