What are you basing your numbers on?

I would say (also going by known and/or rumored specs):
Wii U = 1.5-3 PS360 (but this one is also still very vague as we don't know anything about Clock Rates yet)
PS4 (A10 rumor without dedicated GPU using an unmodified(!) HD 7660D @ 800MHz*) = 2-2.5 PS360
PS4 (A10 rumor using an additional dedicated GPU at least as powerful as the HD 7660D) = 4-?? PS360
XBox720 = not enough known in my opinion to say anything

And all of that is ignoring that the new consoles are built on a more modern hardware!

*the HD 7660D has ~600GFlops while the XBox360 GPU can do ~240GFlops. And I still think you can modify the HD 7660D. OC has been done to ~1200MHz which makes the GPU around 50% faster in theory if you provide enough bandwidth. And AMD can do more than just increasing the Clock Rate. The CPU is about on par to the PS360 ones but that strongly depends on how you use it (highly optimized Cell-Code probably would still crush this CPU, but that kind of optimizing isn't really possible in games afaik. In real world gaming the A10 may well be faster on average).

Yes, I find it strange too, considering that A10 GPUs (which are VLIW4) can be crossfired with VLIW5 hadrware - so why not with GCN? I do admit I don't have required knowledge to fully understand all the intricacies of CrossFire, but I would assume they could make those 2 GPUs work togethr (now when I think of it, wasn't there quite some time ago a rumour just about that, using some different approach in pairing GPUs?)

You are mixing up too many things. Firstly, the A10 APU dram clock is limited to 1866MHz. This has nothing to do whether faster drams are exepensive or not, that is just the design limit (the higher the clock in a chip, the more engineering trouble. So there is essentially a point where more clock is just engineering suicide). Secondly, the A10 APU has two 64bit ddr3 memory controllers (so you can't hook ddr4/gddr5 dram to it to make it faster).. That gives you a 128bit bus width, no more, no less.

But can't AMD modify the memory controller? Without doing that of course there won't be a higher bandwidth possible. But as GPUs all the time use different bus width I thought that changing some parts will also be possible in the APU.

And 1866 is only the supported maximum Ram frequency. Tests with 2133 have been done with some mainboards that allow it and the FPS in games profited from it (but only by a small amount when not also OCing the GPU in the APU at the same time).

You are leaving out a few key facts here. For example, the WiiU uses an average of 40W according to Nintendo. That translates into a clock rate for the CPU in the 2.2-2.4GHz region. There are other hints like the fact that some (launch) software is NOT running at 1080p, indicating some CPU starvation. So CPU wise, the performance of the WiiU would only be around 60-80% of a PS3/XBox. However, the heavily modified GPU (I don't think there is any similarity left to the standard ATI 4xxx GPU) makes good for that so WiiUperformance (Whichever way you define "performance")  is around 130% of the PS3/XBox as I explained in a precious post. Also my previous estimates for PS4/XBoxNext are still reasonable, based on what we know at this time (and there are enough hints for the XBoxNext). SImply forget that 4-? times more powerful (again the meaning of "power" should actually be defined).

I'm just using maximum Flops as a measurement for power for the GPUs (ignoring Texel/Pixelfillrate) and for the most part just ignored the CPUs. Only changes I made because of the CPUs: I set the lower boundary for the PS4 without a dedicated GPU to 2 instead of 2.5 (edit: and for the PS4 with d.GPU to a 4 instead of a 5) and reduced the minimum of the Wii U to a 1.5 (instead of a 2 I would have assigned otherwise, because I assume the WiiU-CPU needs some GPGPU-help to do on par with the PS360-CPU).

Some comparisons (I left out memory bandwidth, too much unknown there):

Xbox360:
Pixel: 4 GP/s
Texel: 8 GT/s
GFLOPS: 240

WiiU (based on e6760 rumour):
Pixel: 4.8 GP/s
Texel: 14.4 GT/s
GFLOPS: 576

WiiU (RV740 speculation @400MHz and @600Mhz (Mobility 4830 equivalents)):
Pixel: 6.4 GP/s ; 9.6 GP/s
Texel: 12.8GT/s ; 19.2 GT/s
GFLOPS: 512 ; 768

A10-5700 (7660D@760MHz):
Pixel: 6.1 GP/s
Texel: 18.2 GT/s
GFLOPS: 583.7

A10+7770 (@1GHz):
Pixel: 22.1 GP/s
Texel: 58.2 GT/s
GFLOPS: 1863.7

Just as reference - 7970GHz edition (with boost):
Pixel: 33.6GP/s
Texel: 134.4GT/s
GFLOPS: 4300

Of course AMD can change things. Makes the chips more complex and bigger (translation: a lot more expensive). The point, however, is not to design a console with world-record-memory throughput, the point is get to 1080p/60Hz as cheap as possible.

There is a point, though, when your cache controller outfoxes your memory bandwidth (basically meaning all the data required by the CPU is in the cache, no matter how wide your memory controller shuffles data). On the PC, that bus size is apparently 128bit. Then you can play with more channels.. but it gets more expensive. With APU, things get even more complex as the bus width is doubled inside the chip for the GPu part. (Still APU was not a good GPU idea for a console that strives for high graphics performance).

My prediction is 6-9x PS360

That is 2-3x Wii U

Are you using just clock rates as the means to surmise performance against the Cell and Xenon?