The reason 3D acceleration tended to create "sameness" is that much of the graphics work was handed over to the chipmakers, not the programmers.
For some time the technological understanding of 3D acceleration made it necessary for everyone to do things pretty much the same way, from art to scene rendering. There wasn't much room to deviate from standard, simplistic multi-texturing because that was the only thing that benefitted from acceleration. Doing much more than that would cause an unacceptably slow framerate or would simply not be accelerated. So, for some time, people were forced to try to wrangle the most distinction they could by using the same old techniques.
Now, with the power to do more than simple multitexturing, to do more complex calculations, and to add dynamic elements to a scene on a pixel and vertex level, developers are not constrained to do everything the same anymore. The artistic vision can be more fully realized with fewer constraints. This power is the result of finally having enough power to do as many passes on a pixlel in a scene as need for the effect, plus special features like DOT3 blending and cube maps becoming a viable real-time capability.
But still, as great as the results obviously are because of these few key advances, this only lifts us half-way out of the old school mire. The engines that come after DOOM 3 and HL2 will probably make use of the full programmable pipeline in modern 3D cards. There is really no practical limit to the types of calculations you can do. We will see a full range of surface effects that make leather look like leather, aluminum look like aluminum and skin look like skin.
Already the DOOM 3 engine, for instance, goes a long way toward this with its specular maps. Talented artists can use the DOOM 3 engine to "mimick" a number of materials by carefully calibrating and mixing texture data with specular and bump data. But although this is very effective and flexible, I think the next generation engines will even BLOW DOOM 3 OUT OF THE WATER. Instead of having the capability to "mimick" leather, rubber, milk, blood or brushed metal indirectly with talented artistic "fiddling", you can have a precise calculation that actually mimicks the physical properties of the material with light equations designed for that material. Then its more a matter of turning some knobs and pushing some buttons and letting the engine depict the material, rather than carefully manipulating it artistically.
The possibilities will be endless, and it will again come down to the richness or poverty of artistic imagination and production design, not the limitations of technology.
This comment was edited on May 20, 17:03.