Ok, now the results from the master build:
Things I reported on LM, MP4, MP5 and MP6 were almost the same on the build.
Mario Kart - Double Dash: Now goes full speed on game mode select; small speed up on "Records" section on the main menu.
SMG: There were some performance improvements like on delroth´s build, but not in the same magnitude (from 16 to 20 vs. 16 to 23-24 FPS). Got half of the improvement on master, more or less.
Will also bring the FPS log from master.
EDIT: Now with the log (see attachment).
Ok, at this point I'll just ask you to stop posting here because you have no idea how to test things.
http://dl.dolphin-emu.org/wips/delroth-dolphin-flags-opt-4.0-1777-x64.7z
New build to test for performance improvements. Turns out I completely missed one of the main optimizations, thanks to calc84maniac for pointing out my mistakes and assisting me in the implementation. I'm expecting this should yield better improvements than the previous version.
Compare that to the same master version as mentioned in the first post. Please do not post any more results with the old version.
Now it's noticeably faster. in CPU heavy situations, sometimes up to 10% faster. Of course, N64 Virtual Console games get the biggest improvement, nearly 12% faster from the latest build vs the "off" version.
Wind Waker gets a healthy 5% boost during interiors (where it's not GPU limited) and tends to keep up higher framerates running around as well.
Metroid Prime 1 seems entirely GPU thread limited; this doesn't have any affect on it.
NBA Live 2006 usually requires an overclock to run full speed on my computer, but this just pushes it over the top. It's hard to give an exact amount because of how the game is; tons of lag/slowdown in menus. But, the gameplay itself is noticeably faster.
It's hard to really grasp the full benefit of the improvement because of how GPU thread limited Dolphin is currently, but it's definitely a fairly good one.
Looks like I got about 8 - 12% (game fluctuates in framerate a lot) improvement in Super Smash Bros. Melee as well.
Wow. WindWaker is much better. Before I could "feel" the emulation happening and this is almost like native.
I'll test more and report back in detail.
edit: I think part of my improvement was because I switched audio backends before testing this. My bad.
With all settings equal I suspect your build is reducing slowdown events. In the previous version I noticed 3 bad slowdowns, in yours 1. Which was during an indoor fight.
In the menus of Timesplitters 3 (MMU Game) is faster, although about 5% faster in game.
Timesplitters 3 - Menus
42/45/43/39 - 78% (M)
49/50/47/51 - 85% (flags-opt)
I tested Call of Duty - Big Red One (MMU Game), the most CPU limited game i have, about 1fps faster, still getting less then 10fps in that game.
Yeah, that won't help much MMU games.
Any MMU Optimizations planned in the future?
I don't know of any. In fact the work around MMU seems more targeted towards making it more accurate and slower (since our current MMU implementation is incomplete and still can't run a few MMU games

).
That's irrelevant to this thread though, please keep the discussion to testing results of this new change. I'd like to know if it's worth annoying magumagu/HdkR to implement it for JitIL/JitARM (since it's required for getting it merged).
In Piglet's BIG GAME (gc) i noticed some missing graphic effects (there is no indication around selectable objects compared to master) and had a couple of unexpected crashes using the latest download.