Just had an odd thought. Thunderbolt is generally too slow to get good external GPU performance, since it's 4xPCI E. But what about using CUDA or OpenCL acceleration? Could you pop a Quadro Pro or Tesla card into a Thunderbolt enclosure and get decent GPU renders from Adobe or blender Cycles? Or would you be better off buying a tiny PC with a pumped up graphics card as a render box over a network? If the former, I wonder if nVidea or ATI would consider making a native Thunderbolt stream processor box.