Kam Leo hypothesized:
You might be able to get improved performance by changing from the kernel's default I/O scheduler (CFQ).
I switched the default to the "anticipatory" scheduler (leaving CFQ and deadline built-in) because of xine lock-ups and that DEFINATELY did the trick (2.6.22.6). I have not noticed a difference otherwise.
In the kernel .config file grep "IO Schedulers"
