Mostly agree with the exception that there is no way they are switching from Intel - if anything I wouldn't be blown away if Intel switched to Macs a year from now...
As for Semi P.A, I think more significantly Grand Central will be restricted to already released intel Mac's as a one-off concession, but going forward's GC will not work on any of the wintel chipsets - hence the 'two bridges' thing again - Apple's own intel hybrid, leveraging a special Semi P.A chipset 'brain' will be required to allow grand central to activate for anything after Penryn, thus delivering STUNNING performance advances for new Mac's (as well as existing ones). Current PC system builders using mainly off-the-shelf parts will simply not be able to match this performance at any price.
You can be sure if Apple have had a parrallel computing breakthru as Steve Jobs has suggested that they will protect this technology going forwards - and the best way to do that is to design your own custom chipset (in partnership with intel).
The signs are all there...
Their parallel computing is likely going to be with the fact that GPGPU progress has exploded in recent years. AMD-ATI for one is onboard with Apple on OpenCL and their 4850 and 4870 cards are producting 1.0 and 1.2 TFlops, respectively. Only issue is that they do not have their GPGPU API down yet w/r/t OpenCL.
NVidia is pushing 933 GFlops per GTX280 with their Tesla platform. Nvidia is a bit further ahead with their CUDA platform.
Microsoft's guidlines with DX11 calls for Compute Shade, which will basically standardize the use of shaders on a GPU to perform GPGPU calculations.
Intel is coming out with their own discrete GPU, Larrabee, which will run mini-x86 cores in parallel and will be > 1 TFlop in power supposedly closer to 2TFlops.
If anything, all this talk of using parallel computing is just Apple going along with what the big CPU and GPU makers are already pushing. Adobe and other people are already onboard with this (see folding@home as an example) and even if Intel doesn't want GPUs to overtake their CPUs, they already know it, hence Larrabee.
And you sure have a lot of faith in what this custom chipset can do for Apple. You think Intel is really that stupid to let it happen? Intel makes the vast majority of its money selling CPUs and associated chipsets. And the vast majority of those go into systems running Windows of one form another (be it home, enterprise, or server). Intel switching to Mac and optimizing for Mac is basically shooting itself in the foot and putting everything on Macs and limiting itself to Apple since only Apple offers its OS. In other words, they would put themselves at the mercy of Macs.
It's not going to happen - they aren't going to allow their CPUs to be run on a chipset that is much faster than their own chipsets, thus cannibalizing possible sales to other OEMs that bid against one another, thus pushing their own margins up. If anything, they'll pull the plug on CPUs for Apple rather than allow another chipset to beat itself thoroughly. At most, they will allow optimization chipsets that boost performance but do not overwhelm their own setups.
Intel is much like Microsoft - they're not going to jump to Macs or other hardware platforms. They have the muscle to set the standard (x86 for example) and wil continue to do so (as they pretty much said no QPI for Nvidia until Nvidia basically said SLI will be allowed on the X58).
Furthermore, if you think Apple is just going to forego the CPU - forget it. The CPU still has the instruction sets for complex calculations no GPU has yet since GPU's are very specialized still in their coding purposes. And if they abandon x86, kiss goodbye to any ability to run Windows parallely.