Tess -
Nah, it's not just positioning. Lots of things work great in the lab, or in $25,000 workstations where people buy $2000 annual service contracts, but don't work in mass produced desktops. Heck, we could increase x86 by 30% or more clock speed right now (ignoring the reliability and damage issues I already mentioned) by water cooling. Water cooling has been available forever. No one does it (other than a few gamers) on the desktop because it's expensive, unreliable, loud, etc. Peltier junctions are another thing people could use. Expensive, etc. Aerosol cooling is another one. DEC and Sun did it in the 1990's. It's gone nowhere.
And, as for on-chip optimizations, I again point out that RISC is easy to do low power, but you pay for it in performance (that's why everything uses ARM, etc.). I don't care what clock speed powerpc runs at because they have a far easier problem to solve. They assume the compiler is doing a lot of the hard work. In CISC, even though there's a RISC-like thing there doing the computations, there's this whole chunk of logic that has to behave like a compiler does for RISC. It also means that you end up having to deal with all sorts of weird addressing modes, self-modifying code, etc. which means all sorts of additional hardware critical paths. Saying that RISC and CISC are the same is like saying that a Lotus is like a Porsche. Yeah, they both go fast. They both use some of the same fancy engine technologies. But there's a lot more going on in the Porsche that has to be dealt with, including about 500lbs of weight.
x86 clock rates will slowly climb, particularly "phony" clock rates ("turbo," etc.) The actual, average, per-core clock rate will slowly climb, but it will not hit 5GHz any time soon (in several years, of course it will).
AMD will probably go a different route. 2-4 x86 cores running around 3-4GHz, and a bunch of special purpose RISC-like cores running faster as needed.
