As the xf86-video-ati radeon driver doesn't provide a textured XVideo adaptor yet, I assume
you're referring to using an OpenGL output plugin in your video player. The reason this
performs so badly is that the Mesa r200 driver is very inefficient in getting texture data
from the application to the GPU, involving several memory copies and possibly other
A textured XVideo adaptor (as implemented in xf86-video-intel e.g.) shouldn't require
significantly more CPU cycles than an overlay.