In my humble opinion, I believe the latest graphics processors, such as the R300 and possibly the NV30, would definitely be able to push more polygons than the Doom 3 engine is going to output while running the somewhat complex shader routines. Even considering the shadow volumes, these processors with ~4 vertex shading units each, who have hardly been tapped, alongside their nifty programmability, should clearly run doom or any game with more polys than they are currently exposed to without having to dumb down effects or resulution.
We should never pass judgment to quickly on today's latest hardware while running games that have been developed to run on the lowest common denominator. I am sure much more could be done if Carmack soley targeted the R300 and the NV30 (which he eventually will). So to say that high polygon games with complex shaders is out of the question for this generation hardware is most certainly wrong in my mind. How do we know we'll have to run Doom 3 at 800*600 for respectable framerates when, with good coding (a definite for id), the engine should not choke the R300 or NV30, even if it carried more polygons per model or if it had a resolution of 1024*768. Being that the two processors (or at least 1 that I know of) handle 2-3 shading ops per cycle, per pipeline, with 4 vs's and 8 ps's, they should be more than satisfactory for running the game at full settings (assuming cg, or some other minimal hardware optimization is made).
Aren't the hardware engineers the ones who set the boundaries of a rendering engine's possibilies. The boundaries of Doom 3 are set right around the first generation geforce line, with some optimizations for the more modern graphics processors.
P.S. I'm not sure if I went a bit off topic, but it truely buggs me when the latest hardware is shot down over a game engine which was not comfigured specifically for it.