I went to the training dummy in Org to test out stuff. With about 150 hit, I very rarely miss against raid bosses, at least WotLK bosses. Against the raid boss dummy, I missed so much I stopped even trying to test. Instead, I tested against the level 80 one. That may have skewed the results a bit, but probably not much.
I would have done more testing, but my food buff fell off, and I didn't have another food with me that had the exact same spell power.
Couple of random things I noticed:
Ran out of mana fastest with Wrath spam. No surprise there.
Starfire spam lasted the longest.
Crits really do return a good amount of mana.
Crits add a TON of DPS. Not just the crit amount, but the .5 second cast reduction on the next spell.
Here are the numbers: (most tests run twice, casting until OOM)
ROTATION - DPS (Eclipse procs where applicaable)
Wrath spam: 2311, 2289 (AVG: 2300)
Starfire Spam: 2217, 2345 (AVG 2281)
Eclipse: Wrath spam then Starfire: 2319 (2), 2354 (2), 2370 (3, last one at the very end, with 2 Starfires) (AVG 2348)
Eclipse: Starfire spam then Wrath: 2361 (3), 2321 (3) (AVG 2341)
Then for grins, tried the trusty old rotation. Moonfire/Starfire: 2314, 2530 (AVG 2422) Don't compare this to anything but the Starfire spam numbers. Had I done Moonfire with the other rotations, they probably would have seen a similar DPS increase.
What do I make of these numbers?
Running a test 2 or 3 times is far from a super solid measurement. However, since the ranges between tests are fairly small (in most cases) these numbers do indicate a few things to me.
* Moonfire sure does add DPS! No surprise there. Would have liked to try with IS too, and Moonfire/IS with the Eclipse rotations.
* As expected, Wrath and Starfire are pretty close in DPS. However, Wrath drains mana quite a bit faster.
* Both Wrath and Starfire spam ran roughly around 2300 DPS. Both Eclipse rotations ran around 2350 DPS. Eclipse proc watching added something in the range of 50 DPS. I had hoped for quite a bit more. On longer fights, it may show a little improvement.