But how soon would mainstream be & mainstream would suggest to me not extremely costly. I would say that's at least 2 years away at best on a GPU. Its great to have the slot available to develop the hardware for it but its not going to be mainstream any time soon.
Wouldn't it seem more practical to use multi cored CPU's for physics? Like Quad cores especially which would be mainstream sooner then a GPU solution wouldn't it?
The processing power is already there to handle physics processing, even in last-gen mid-range cards. It's simply a matter of implementation. Games need to start taking advantage of GPU physics through middleware like Havok FX in order for NV & AMD to sell physics-branded cards.