My mode of working is to have something---typically mplayer h.264 or mythtv---running on the left head while I work on the other three heads. Unfortunately, all the video options currently available for my setup are sub-optimal.
First of all, I must use Xinerama rather than TwinView because I have four heads and want them handled logically separately. E.g., I want a maximize operation on a window to maximize to one head, not two. This means VDPAU is not available, so I am forced to use either OpenGL or Xv.
OpenGL: Has frame-dropping performance issues when I do anything non-trivial on the other heads. E.g., simply scrolling through the available folders in Thunderbird causes the video to become choppy and drop frames. Sometimes, it's as hard to track down as Firefox getting into a mode in which it causes the video to play *very* slightly unevenly... a dropped frame or two every second. This is the *best* of the available options as far as I can tell.
Xv: Causes X to consume 100% of one core, and results in any X operation being unresponsive. Scrolling through the list of folders in Thunderbird might, for example, take 10 seconds. Unacceptable on a I7-920. This might go away were I to disable vsync, but then I get tearing, which is even more unacceptable.
What am I to do? This is a killer setup, but I lack the software to take full advantage of it. What exactly is the technical problem with getting VDPAU support on Xinerama? I understand that this is probably difficult with the resources NVIDIA is willing to commit to Linux. Therefore, insert rant here about opening up more of the source code to the free software community, so more people closer to the problem can work on it.