Will it beat my old geforce 2 ti for 3d performance?
Would it be possible for one of these cards to accelerate cedega by doing d3d->ogl transformation in hw, and then feed it to one of these high end nvidia ones?
It's hard to predict exactly how the open graphics chip will perform. The plan is that it'll produce a maximum of 400 million pixels per second, but since that isn't the maximum memory bandwidth, it'll continue to perform the same, even as you turn on bandwidth-hungry features, which means it should actually perform very nicely for the mid-range sorts of applications.
As for converting d3d to ogl, this chip is a 3D engine. It's not programmable in the way I think it would need to be to translate commands from one API to another. I suspect that would always be done best in software. However, there's nothing stopping them from writing a native driver for the open graphics chip that uses the d3d API.