Not really. Many things are still done in Engine, especially lighting. To have some bumpmapping effect I'd need to pass over lighting information to the renderer to even realize it. Not even that it would be still slower then and would require even more performance, it would mean to do a lot of things "manually by hand" which otherwise the GPU and some routines would do almost entirely alone.Good to know. And why is it not feasible? Does the code abstract the D3D/OGL functions too much? And what do you mean about GPU acceleration, doesn't the renderer automatically utilise it when those functions are used?
Lighting should be done nowadays in GPU, its much faster and to realize things like bumpmapping and realtime shadows (really lighting dependent shadows then) would be way easier.


