[TheFluff]

Some pseudo-scientific tests (more scientific than looking at jpg's of graphs anyway)... This time I'm measuring CPU time. See below for test details.
MPC - first 120 seconds of F-B Mahoraba 12 (five tests):
1. 30 seconds
2. 29 seconds
3. 28 seconds
4. 29 seconds
5. 29 seconds
Average CPU time: 29 seconds
Software: MPC 6.4.8.7, ffdshow 2005-12-20 CCCP build (for both video and audio), Microsoft's builtin AVI splitter.

VLC - First 120 seconds of F-B Mahoraba 12 (five tests)
1. 20 seconds
2. 19 seconds
3. 20 seconds
4. 21 seconds
5. 20 seconds
Average CPU time: 20 seconds
Software: VLC 0.8.4a.

Zoom Player - first 90 seconds of ZX Starship Operators 09 - five tests
1. 53 seconds
2. 53 seconds
3. 52 seconds
4. 52 seconds
5. 51 seconds
Average CPU time: 52.2 seconds
Software: ffdshow 2005-12-20 CCCP build, ZoomPlayer 4.51 standard, Haali Media Splitter 2005-11-25.2, VSFilter 2.37.

VLC - first 90 seconds of ZX Starship Operators 09 - five tests
1. 51 seconds
2. 49 seconds
3. 51 seconds
4. 50 seconds
5. 51 seconds
Average CPU time: 50.4 seconds
Software: VLC 0.8.4a.

The test was conducted with a fresh, reset-all-settings install of CCCP. Some stupid ZoomPlayer defaults were changed (hide the control bar, autosize interface to fit video size, space bar == pause video), but otherwise all defaults. Task Manager was kept running in the background. The player was started with the video and left alone for the duration of the test. When the end point was reached, the player was paused and the CPU time for its process noted. The player was then closed and the process began anew. The H.264 test file had softsubs, but they were not visible during OP, since it was hardsubbed. Hence, VSFilter was loaded but not doing anything.

Speculations: It's weird that VLC is that much faster on XviD. After all, the decoders are the same... In the H264 case, I'm putting the difference off as a result of Zoom's slower loading time (since the player start time is included in the test) and fancier GUI, plus the extra effort to create the DirectShow graph.

Incidentally, some casual testing with ZoomPlayer indicates that it actually uses slightly less CPU time than MPC for the Xvid test. For the H.264 test, there's no noticeable difference between Zoom and MPC.

I also tested MPlayer a bit, and found that it's about as fast as VLC, but with HALF the RAM usage, and this is despite not using any SSE/SSE2 optimizations (my CPU has them, but MPlayer doesn't want to detect them, it seems).

Concerning the weird VLC stats of some people: I think the measurement has gotten lost somewhere in some thread. I have no idea why it behaves like it does, but I'm not believing those results until I can reproduce them myself (which I can't).

BTW, @subcool: yes, you can set that value negative with the GUI. Try it for yourself. Just press the down arrow button and watch it happily go from 0 to -1...