Kidding aside, this doesn't make much business sense. A lot of people buy Macs and run Windows alongside OS X because their work requires Windows. Why would Apple remove that whole segment of the market from their purchasers just to save a few bucks on the chips? I don't think they will.
I also think this is slightly unlikely because it would probably require substantial investment in rewriting a lot of the plumbing for OS X (I don't know much about it).
And even further, why would they work on building their own chips like the A4/5 and then put in the effort to end up using someone else's? If anything, they'll either switch to something else just in the MBA's or you'll see A* chips in there or something.
I guess that this would make sense if they started making laptops running iOS.
Mostly it bugs me that ARM is seeing more and more use yet I still can't drive up the street and buy one on a motherboard. Freezing out hobbyists is an alarming step backwards for the industry.
Moving away from the X86 instruction set 5 years after the move PowerPC is NOT going to happen, period. ARM is fine for the iPad, but for high performance work it doesn't cut it.
ARM tech & Intel have proven that they can scale, but only Intel has the performance advantage for years to come.
And come on! Fragment the Mac market in laptops with ARM and workstations with Intel chips is a really bad move. The only laptop that would have an ARM chip would be an iPad with a keyboard.
x86 will have have lasted 8+ years at Apple, which I think is reasonable when you consider how much Apple products have changed since x86 was adopted.
I also think a RISC instruction set is starting to look pretty nice, since (as I understand it) x86 chips break CISC instructions down to RISC-like micro-ops anyway. Why not move that process off the die (where space, power and thermal requirements are strict) into the compiler.
The transition away from PowerPC wasn't that tough in reality. If anything, it told Apple that such a change is very possible. With Windows 8 coming to ARM, it's not that big of a deal for Apple to follow suit.
But this was necessary to achieve any kind of real performance with x86. It's a terribly-designed architecture where all of the instructions are scattered all over the map. There's no unifying design, and most operations that should be simple to implement require scads of silicon. Much of the power wasted on x86 chips is in that instruction decode hardware. It's terribly inefficient. (It's also a horrible instruction set to deal with.)
ARM competes because of a more power efficient architecture; however, this difference is less of an issue if Intel's fab process can get farther ahead of the competition.
In short term(next 4 years) Intel's lead fab lead will likely continue to grow, because Intel switched to Tri-Gate(FinFET) at the 22nm node while TSMC and others will not adopt similar technologies until 16nm node(2014 or later).