This was a very interesting article to read and also very insightful even though I don't agree to all points in it. I really wish we had more articles with this quality level! Thank you.
You mention Itanium which is interesting but I don't think it is fair to mention Itanium without mentioning for example SUNs architecture. I hear a lot of talk about how good SUN Servers are and how little performance/cost you get from IBM Power6 while I hear no word about Itanium in those circles.
I would rather say that the problem game developers actually are facing today is that most people don't have the latest hardware and therefore they have to reduce the hardware requirements so that the released games can be used by a broader audience and not just by a smaller group of gaming enthusiasts. So the challenges they are facing are the challenges that come with producing games that work on weaker hardware but at the same time utilize the additional performance gained from newer hardware.
I've always dreamed that one day there will be computer games with high terraform details, with trees and foliage with excellent detail, high variety of species and physical response to elements such as characters passing by, the wind in my face or a blasting shell. When I jump into a lake I see how the water splashes realistically and interacts with my character and others that are in it. See the following demo for some examples:
http://www.youtube.com/watch?v=2_RSYoCEMF0
Moreover, there are a bunch of more realistic but very computationally intensive rendering techniques than traditional raytracing; take for example Splutterfish
http://www.splutterfish.com
which uses a technique that simulates a stream of photons in the rendering process. Effects such as ray scattering, shade diffusion, and refraction from transparent objects comes "naturally" from this process without the need for application of additional effects and the results are incredible.
I also suspect that in the future the 3D models will achieve such a high level of detail that effects such as bump mapping may no longer be necessary. Objects may not be created from scratch in modelers but rather be scanned from real-world objects.
There is much more development to do when it comes to advanced A.I. and imagine a game world of hundreds if not even thousands of intelligent characters interacting with each other, forming teams and doing all sorts of intelligent behavior. That would be immensely computationally intensive.
So I would say no, we are nowhere near a maximum when it comes to the need for additional computational power which I would say is outright insatiable. I strongly disbelieve that programmers have some kind of pressure due to their "inability" of implementing the additional performance granted by those newer lines of GPUs in forms of budget constraints or knowledge based limitations.
There is an insatiable need to be satisfied and I wish the hardware developers all the best in their development of future hardware. And I hope they will still be around competing with each other intensely while making us gamers happy with even better and better GPUs for years to come.