Steven Den Beste has a lengthy exposition of why moving the Macintosh user base to an Intel (or AMD) CPU is exceedingly problematic. I think he’s mostly right, but I don’t know if software emulation of the PowerPC instruction set would be as slow as he suspects, at least from the user’s perspective: the most CPU-intensive task most Mac users seem to engage in is moving around their pretty OpenGL-rendered windows, and even my low-end 300 MHz G3 at work keeps up fairly well with OS X 10.2.
I don’t think reasonable-speed software emulation alone would prompt most users to switch; Apple would have to “add value” over a straight x86 box running XP Home. And, IMHO, the best way to add that value for the Mac userbase would be the ability to run Win32 applications natively—that is, to sell Apple’s x86 boxes as the universal end-user platform, able to run Windows, classic Mac OS, OS X, and Linux applications on one desktop.
Could Apple have such a project secretly in the works? I don’t know, but they’re essentially 3/4 of the way there already with the rootless native OS X X11 server (presumably you could run LinuxPPC code under Darwin-PPC/OS X with a thin emulation layer, and likewise for Linux/ia32 under Darwin-x86), and if a few dedicated hackers can produce WineX or CrossOver Office, I’d imagine Apple has the resources to do the same, perhaps building on the work of one of those projects.
Another question is the business model—does Apple stick with semi-proprietary hardware, or go back to licensing the operating system (perhaps with a boutique branded hardware line for the Alienware set and hardcore Mac fans, returning to the NeXTstep model)? I think the latter plan is more viable (mainly because I can’t see much value in building a semi-proprietary x86 box when you can get off-the-shelf hardware much cheaper), but the former avoids the very real problem of trying to directly compete with Microsoft on the desktop and losing the native Office port, although its value might be overrated with good binary emulation of Win32.
Sean Jordan has responded by email, noting that Apple is probably more likely to move to the new IBM PowerPC 970 CPU; despite Den Beste’s cynicism about the viability of the PPC 970, I do think that's the more likely option (although that may be the closet Motorola fanboy in me talking). However, it’s still fun to (continue to) speculate about the viability of an Intel (or, more likely, AMD) move, especially considering that OS X’s core (OpenStep) ran on Intel chips before it ran on the PowerPC. He also points to the consistently-excellent Ars Technica as a good source for watching what’s going on on the Apple CPU scene.