I began my analysis by creating a worst case estimate of the amount of data being sent to the graphics card. First, let's equate 4 video frames to one 1280 X 1024 image. Then assuming 2 X 1280 X 1024 images with 4 byte pixel depth at 30 fps we get 1280 X 1024 X 4 X 2 X 30 = 314,572,800 bytes/second worst case. According to Intel (http://www.intel.com/design/chipsets/pciexpress.pdf) the PCI Express X16 bus is capable of handling 4 GB/s full duplex. Therefore, it doesn’t seem like the bottle neck is in the graphics card. The observation that the CPU is being maxed supports the conclusion that the bottle neck is earlier in the pipeline or possibly due to mishandling of the tasking of graphics card functions. Another, reference point is the fact that there are numerous applications that are being used to modify an incoming DVD data stream for display on high resolution LCDs up to 2560 x 1600 often with additional image processing.
This may all be hot air but this is where I am so far.