The best two T9 pieces are hat/chest, so assume you have those on for your two-piece. The best T8 are legs/shoulders, so let's assume you upgrade those two slots from 226 to 258. You gain:

79 spellpower
43 hit
-64 crit
52 Int
113 Spi

Testing in my current setup (with 2T9 and IMF added in), exchanging those stats for the set bonus trades about 100 DPS for the 43 hit (which presumably you'd cover by swapping other items). Probably a net loss unless you were really struggling for good hit pieces.

For something more hit-negative, let's try upgrading legs/gloves instead:

79 spellpower
-14 hit
-76 crit
76 haste
52 Int
106 Spi

That looks like it loses 60 DPS and 14 hit. Definitely not good.

I'm seeing 2T8 worth about 415 DPS and 4T9 worth about 175 DPS, so this isn't surprising.

Since maybe the T8 pants are weak if you don't need hit, you could conceivably upgrade T8 legs/shoulders to T9 shoulders/gloves and swap the offset piece. This might be the most efficient plan, but for the moment it seems unlikely it would bring any gain, let along a large one.

Numbers would change slightly if I increased the stats to reflect what I might be wearing in late Colissuem, but not by much (esp. since the set bonuses scale quite well).

TLDR: I still find 2T8/2T9 to be BIS.

One's a useless MP5 proc on heals, the other's a lightning capacitor which makes it hard to implement in Rawr correctly without adding code. Instead, I just gave it Lightning Capacitor's proc, which bumped it to the top of the list. It shouldn't change much else, and won't itself change much with proper implementation.

That's not exactly a fair statement to make. There's no connection between the opacity of Rawr's calculations and its inability to allow you to enforce certain gear setups. Rawr is built to do two things: tell you the expected output of a given outfit (for lack of a better, unambiguous term) of gear and, based on those numbers, find the optimal outfit from a given set of available gear. It's built for certain constraints, usually applied to tanking classes, such as keeping resistance, actual health, survivability, etc. above/below a certain value, but it's not designed for such micromanaging as specifying that you should use 2T8 and 2T9 for the simple fact that it breaks optimality. There's a reason to have X resistance, or be uncrittable, or having over Y Health. There's no reason to specifically choose 2T8/2T9 over 4T9 if it's not optimal.

That out of the way, sometimes the optimizer fails*, and this appears to be one of those cases. I marked the gloves and shoulders of T9 as unavailable to force 2T8/2T9 (note that, when I said it was difficult before, I actually meant it was tedious. There are 30 2/2 combinations, and each one has to be forced manually), and the optimizer output a set with 3 DPS higher than the 4T9 setup. I'm reasonably sure that I can coax even more out of it by doing some more work defining sets to optimize, but at this point I'm willing to say that they're neck-and-neck.

*The optimizer is simply an interative process, not an enumerative process. It doesn't try to find the value of every conceivable combination of gear possible from your available set. Instead, it starts at a base point, and upgrades piece for piece until no possible gear swap it tries produces a larger output. Thoroughness controls how many swaps per slot it will try before it gives up, but there's always the possibility that it won't catch something.