0) Original:                           165.67fps
 1) Hack out device_preload_textures    187.08fps       (see 18, reverted. More than expected)
 2) WINEDEBUG=-all                      178.43fps
 3) rt location housekeeping            186.29fps
 4) ds location housekeeping            200.93fps
 5) redundant fbo checks                254.99fps
 6) FBO validation                      293.61fps
 7) mutex call level                    294.08fps       (essentially unchanged, reverted)
 8) SetPrimitiveType checking           297.80fps
 9) CFLAGS="-g -O2"                     297.80fps       (supposed to change nothing)
10) CFLAGS="-g -O3"                     308.05fps
11) CFLAGS="-g -O3 -fomit-fp"           337.11fps       (yep, I know it is -fomit-frame-pointer)
12) compile out debug msgs              341.77fps       (huh, once this ran slower)
13) no -fPIC                            344.21fps       (ok, not that much change here)
14) findTexUnitMap hack                 390.15fps
15) no wined3d mutex                    449.40fps       (unrealistic)
16) no gl locks                         848.77fps       (Santa Claus is real)

---- Notice ----
From here on(and probably an earlier point) the individual items cause way too high
performance improvements. Keep the inverse proportionality in mind
----------------

17) context level counting              887.41fps
18) hack out device_preload_textures    973.35fps       (let's see how much that does without the other hacks)
19) turn render to fbo off              919.39fps       (interesting, reverted)
20) Turn off glsl                       973.35fps       (note: the shader and constants are static, reverted)
21) remove point sprite warning         976.68fps       (margin of errror)