Well, the f-buffer may also be able to remove other limits, such as the number of textures sampled (I don't know about the specific f-buffer implementation, but it is theoretically possible), and the number of temporaries available. It may even be possible to use this for data-dependent branching (it would be very slow, but it may be possible).
Hopefully nVidia has similar technology coming. It really is essential to remove all such limitations.
"Physics is like sex. Sure, it may give some practical results, but that's not why we do it." - Richard P. Feynman