Universal binaries are available for Mac OS X for the last 10 years. Ok, it was called NeXTStep at that time though :-)
So go and google for 'fat binaries' and 'NeXT' to see what is available.
No one prevents you from supplying an app in three flavours:
I guess, for Apple this is one of the few ways to survive. With the nice centrino chpisets, even Win* notebooks are not too bad these days. And they are way cheaper anyway.
The porting bit is not too hard if you ever contributed to an OSS like NetBSD. So I beleive that the Mathematica port only took two hours.
I think that Apple just needs some more time to e.g. offer new iBooks that are at least that performant than the average Centrino one, but still consume less energy.
By the way: when do you really need to care about byte ordering anyway? It is mostly only relevant when you want to access hardware directly (i.e. device drivers) or you want to use some assembler code to speed up compilation or when you design a new lower level networking protocol where you want some well known 'on the wire ordering'.
If you just write normal apps in C, Objective C or Java you don't care at all.
And for the other apps it is not that hard as for example NetBSD or mcntp prove.