They've done it twice, not once. They moved from 68k to PPC in the early 90s, again with full emulation. In that case it was even more extreme, as much of the OS ran emulated in early PPC releases (yet it was still quicker than running on actual 68k hardware!).

You've highlighted the big difference this time. Switching from Intel to ARM would be a step down in performance, or at least not a step up. There's no headroom for a legacy emulation layer.

However, it does work with the playbook of going ever closer to computing appliances from general-purpose computers.

In the beginning there was the motherboard and a CPU. Then, before homogenic PC era, we had dedicated chips that took care of certain operations. The SID chip in C-64. The blitter chip in Amiga. (Can't remember the name, I'm sorry.) Even the x87 math coprocessor in the 386/486 age!

With advent of PC and the megahertz wars, dedicated peripheral chips became less common - except in SoC environment. Where the x86 world went with raw processing power, embedded world had to find ways to fit specialised chips on the board.

My experience is mostly centred around crypto accelerators, but I know from very painful experience that all Maemo devices had on-board DSP units to handle some sound decoding, and pretty much all video processing. So the pendulum swings: CPU for everything, then peripheral devices for specific high-intensity jobs. Some of the most commonly used get integrated into CPU's, making entire classes of chips irrelevant ... until the next CPU-intensive thing comes up, and the main processor is again too slow.

Apple is banking on their ability to both predict and dictate the direction of near-future computing needs. I expect the A7/A10 boards to come up with all kinds of integrated support chips to handle the heavier loads.

As long as their predictions are correct, all is well. Any bets on what's the next CPU burner that will require a dedicated ASIC to preserve even the semblance of battery longevity?

I think it's technically irrelevant, how many people from the 68k/ppc era are still there, and the os/tooling/techniques back then have probably nothing to do with what's in place nowadays.

Pyschologically it still matters though, as you said, they did it twice and the first was surely a harder task.

They almost certainly have some 68k/ppc guys left, and while tooling has changed they at least had a historical recipe (e.g., go ahead an emulate somethings, transition gradually, etc) that was proven to work.