> AFAIK, both Playstation 3 and Xbox 360 devkits are based on Intel
> x86/Windows to be used in combination with the console HW.

Yes, it makes sense to use relatively cheap off-the-shelf x86 computers as part of the devkits even for console hardware with a different ISA, but it surely wouldn't make sense the other way round, i.e. to develop and manufacture special devkits (or even special chips) based on Power Architecture for console hardware based on a different ISA. At least I can't imagine why the devkit for a console to be based on ARM or x86(-64) should be based on Power Architecture.
Besides:
"Before the launch of the Xbox 360, several Alpha development kits were spotted using Apple's Power Mac G5 hardware. This was because the system's PowerPC 970 processor running the same PowerPC architecture that the Xbox 360 would eventually run under IBM's Xenon processor."
http://en.wikipedia.org/wiki/Xbox_360#Development

> But what if the console HW isn't there yet?

Use cheap x86 devkits rather than develop expensive, say, Power Architecture devkits for a, say, ARM console.
This of course means that in reverse, a devkit based on x86(-64) wouldn't imply the slightest about the ISA of the respective console's final hardware.

> *if* the new Xbox would turn out to be yet another fruit of the
> Microsoft/nVidia cooperation on ARM, then the new "killer games" to
> sell the platform needs to be developed *now*

True, but I still think that would happen either on cheapish off-the-shelf x86(-64) hardware or maybe even on specially developed and manufactured ARMv7 hardware (as 32-bit ARMv8 (= AArch32) will be able to execute ARMv7 code as is, and source code migration to 64-bit ARMv8 (= AArch64) should be easy enough), not on specially developed and manufactured Power Architecture hardware.

> it would have to be something different, but preferably something
> with similar specifications (like RISC, many registers, etc).

Admittedly, instruction count and register count do have an influence on the viability of system simulation, but I'm not sure if that would justify the costs of developing (or having developed) what's probably a new Power Architecture chip (provided the rumour is true, that is).

> just speculating here of course and only hypothetical in any way

Agreed.

> fact is that it kind of coincides with nVidia launching its "x86
> killer" ARM chips (sometime in 2013)

Each and every hardware announced for 2013 "kind of coincides with nVidia [supposed to be] launching its "x86 killer" ARM chips" ;-)

> *there is* this statement that "current devkits will not mirror the
> final hardware".

..."in appearance". As far as I as a non-native English speaker am aware, this could as well just refer to how the hardware looks, i.e. the case design etc. And even if not, there're many things that can differ in a hardware design without the ISA or even the actual main processor chip to be any different.

> If *you* would be using the nVidia "Denver" (to be released in 2013)
> in your next games console, a good selection of games need to be
> ready within one and a half year of time from now, a devkit needs to
> be released about now, and the devkit needed a year or so by itself
> to be developed, how would *you* have done it a year ago (or whatever
> time it takes to develop a devkit)?

I'd have opted for off-the-shelf x86(-64) hardware and done with that what you think they have possibly done with specially developed and manufactured Power Architecture hardware.

> I think my question still remains though...?

I do not know enough about modern game development to be able to judge the usefulness of 16 hardware threads or even 16 processor cores for that type of application.