I forgot to mention Nvidia's VDPAU. With VDPAU you can watch videos with very little CPU utilization. I think VDPAU was added starting with the 2XX cards. The latest Adobe flash player now uses VDPAU when it's available.

I just tested VDPAU/flash on an X6 machine that is running High Performance Linpack with 100% CPU utilization on all six cores. The youtube video ran perfectly. Xbmc also supports VDPAU.

The only way you can offload video with ATI cards under Linux is with Crystal HD add-in cards.