It's completely feasible. The problem I see is that mapping 64-bit addresses into a 32-bit space is expensive. I can't think of anything in ARM that will make things more performant at the moment, but I'm sure there's some sort of way to pull it off. If OS X binaries are relocatable by default, that'd be an easy solution.
Apple has the expertise to be such a client. They're a very significant customer of ARM, and also (unless things changed when I wasn't looking) a major shareholder.
Emulating intel code on the ARM is feasible, though going to be 5x-10x slower for high performance compute code. However: most apps are not high performance compute code; and given that native ARM for most apps is not a lot more than a recompile away, that might not be a problem.
(With notable exceptions like Adobe - I'm sure Apple can solve these individual problems well enough to make that into a non-deal-breaker)
There's only one way to go thinner, smaller, or lighter on their laptops at this point, outside of some huge breakthrough in battery technology.
What doesn't make sense, though, is apps (built for x86)... Though an app-store-only model (and requiring ARM builds for inclusion in the app store) is not entirely out of the question, iOS style. Back to the Mac, indeed.
I don't know if an x64->ARM translation layer for the transition is feasible, performance-wise, though - perhaps someone with more knowledge of the architecture can chime in here.