The ATI gpu client runs off CAL, and the Nvidia gpu client runs off CUDA but not Physx. Nvidia cards have a built in physics processor, ATI cards dont. They simply run it through software, which makes it much slower (Havok Physics).
Even if they made a gpu client for ATI cards which runs off CUDA, i doubt it would be as fast due to how the shaders work between both cards.
ATI cards have more shaders, but the Nvidia shaders run faster which is why they fold faster.
The new projects change things a little bit, but Nvidia cards till come out in the lead.