If they use both TSMC and Samsung, it will not be for the same processor. For all we know this may simply mean Samsung will keep supplying A7 & A6, with TSMC handling the A8.
Different fabs have different process, even if they're at the same node. You can't take the same blueprint to two different shops. Apple would have to essentially design two chips. They would have different performance and power consumption. It'd be practically impossible to use them interchangeably in a single design.
Now they may choose to do A8X with Samsung and A8 with TSMC - still a pain, but at least those are two different SoCs where presumably the iPad one could tolerate an older 28nm process from Samsung with a higher power/thermal profile which by that time would have high yields and be offered at a competitive cost. It would provide some insurance policy to Apple in that the overall demand for cutting edge 20nm chips from TSMC would be eased, while iPad chips supply becoming essentially risk-free.
Not sure where Samsung is wrt 20nm, but there's been no indication of it being ready anytime soon. On the other hand I believe TSMC will have mastered 20nm volume production by this time next year just in time to fulfill the next iPhone's demand. I'm sure Apple thrown their money at TSMC to make this happen on schedule.
I'd agree with much of what is said here. There seems to be too much of an assumption that a design can simply be poured from one fab process to another.
Let me throw out an alternative scheme which fleshes out what you said above.
Rather than having the A8X and A8 be essentially the same chip, but one is fabbed on 28nm and one on 20nm, how about they're actually very different chips, just with the same name.
The A8 is a die shrink of the A7. It gets the benefit of the lower power and higher frequency from 20nm, but it exists basically as Apple's learning how to move their design to TSMC 20nm.
Meanwhile the A8X is a new microarchitecture still on 28nm. The changes that Apple has lined up are fairly obvious. They are likely going to switch in their own GPU, and they are likely going to tie that GPU as closely as possible to the CPU (ie shared virtual memory space). Which in turn means that at some point they probably want something that looks like Intel's desktop designs, with a small L2 (256kB/core, not the current shared 2MB), a large L3 shared with the GPU, and a ring tying everything together. (Maybe they can hit the speeds they want with the current fabric, but I assume if Intel went with a ring, that shows a switch in the middle is just too complicated and slow or hot.)
There's also the move from two to more cores. (While the hard problem is moving from one to two cores, two to three is still not trivial [synchronization changes from "me vs the other guy" to "me vs WHICH other guy"].) It's possible we see three cores with the A7X, but if not, I'd guess we'll see them on the A8X.
Finally there's the dropping of support for ARM32. The point of dropping this support is that it removes a number of operations that limit the maximum speed (eg the addressing modes, the complex THUMB decoder, the shifts and predication for every instruction) and so allows for ramped up frequency. If Apple did this, they might be able to get the same frequency for the A8X at 28nm as the shrunk A7 (rebranded an A8) at 20nm.
Point is: we've all assumed Apple will follow an Intel tick-tock strategy over the next few years ---new CPU design in odd years, die shrink and process upgrade in even years.
But Apple may feel that they can or want or need to move faster than that by running the tick and the tock in parallel on different chips. The people who follow this sort of thing claim that TSMC not only has 20nm up and running, but expect it to have 16nm FinFET ready next year, and presumably Apple also wants to be ready for that...
