its not that the TMU's are slowed down, also look at the 4850 reviews the core hasn't been crippled from what we know so TMU's are in line with what we expect. But you have to consider a few things doing with optimization and die shrinks first. First of all as stated it not an automatic power saved and 65 to 55 isnt that big honestly. Because the number of shaders went up and so did the TMU's granted but look how the core was arranged and the memory sub system the GDDR5 runs alot cooler and uses less power than even GDDR4 does and with you have 8 chips that helps lower first of all. Second of all the core was arranged better this round likly to cut heat output from the core which adds the bonus of less energy expended on heat so you save more power because of less friction in the interconnects to generate heat which is simply a loss of the cards power to friction. All this play's a massive role in the power usage, along with other things such as quality of componets used ect, even the caps can effect the power usage. I honestly belive the power envolope we see has more to do with design that process
Look, that the power efficiency and many other things increase as the fab process matures is a fact, finding the right type and amount of doping, better ways of getting rid of the remains and the time and the way the chemicals are applied are only some exemples of the things that advance and that have to be tested for every process (EDIT: and chip) as no two are equal.
Don't talk about GDDR5 here because it doesn't use it.
There's no doubt all the things you mentioned are true, but IMO the process is a lot more relevant. Ati is always the first to jump to those new processes, so they are simply not mature when they first use them. They take a risk and sometimes it pays off and sometimes not. Not to say that what you say it's also true and, in fact, many of the enhancements you mentioned have a lot to do with the fab process, because being smaller, the arrangement has to change. This is true, but if Ati wasn't able to come with the right arrangement in the first place, Nvidia will probably not do it better. Only that is one reason not to change. Let's see another one:
Price - What I'm going to say can be applied to GDDR5 too. When talking about the price of the fab process and the benefits you are going to obtain, you can't simply compare the price as it is now (or was when Ati started using it) and say it pays off to do the change. Nvidia has 60-70% of the discreet market share, that means it has to produce twice as Ati if they want to mantain it. Current manufacturing costs of 55nm (remember GDDR5 too) are based on the demand of pretty much only Ati's needs and they strugle to meet demand. Mix Nvidia in and you need to multiply the supply by 3x. They can't do that, so prices would skyrocket. As a fact TSMC already upped all of his prices because of this same reason and increasing costs (that at the same time are related to the demand too). If prices went up it would be bad for both companies, not to mention consumers, but because Ati is targeting a lower priced market that situation would be beneficial for them, but in no way for Nvidia. For Nvidia is just better to not enter that price/supply war and make prices lower as long as theay are competitive. Don't dare to ask if this point is not well explained.