Each arch by each respective company still favors certain things, such as texturing etc, as the design allows for certain strengths, as well as certain weakenesses, and is why certain games favor 1 over the other, and this wont change completely. Designing a new arch is a crap shoot, as each design is to last for several iterations, and as an example, tesselation, with it comes certain demands. As devs get into their games, they may favor certain aspects of using the tesselator, well, say youve developed your tesselator 1 gen back, but dont have a new arch. If youve "guessed" wrong in the needs of using the tesselator down the raod, and your arch hits a weak spot the devs are primarily using, such as the case with the 2900s shader resolve, you have to either make changes within your current arch, as ATI did, and beefed it up, or come in with a whole new arch, which isnt possible unless its ready.
Remember, consoles do have tesselation, so that will be here. The lighting effects and tesselation are the prime suspects , plus the optimisations of say DX10.1, which allow for easier dev time and better usage of our compliant cards. XPs down the road, itll be 2 gens old soon, and DX9 will have run its course. nVidia has the potential to emulate certain aspects of DX10.1 currently in thier current cards, and with DX11, therell be more of a unification of this, so designed appropriately, DX11 cards should be very flexable due to the changes in DX11. Make sense? In other words, HW wont be as hamstrung by having to match exactly the current DX model, as long as the card is within that model from the beginning, thus a DX11 card , making it more flexable. Future iterations supposedly wont require newer HW, but Im not sure were there yet. As someone said, will there even be a DX12?
PS I just woke up, so I hope this makes some sense heheh