"All this IMHO for a small benefit: to be able to develop i386 packages on a x86_64 host. How big is the number of users doing that? And does it really work in practice? A lot of configure scripts and apps (including rpm) in my experience seem to get confused if you try to compile something for i386 on a x86_64 host (even when remembering to use setarch). So it's really the best and the safest to use a chroot (e.g. mock) for this purposes as far as I can see.
"So in other words: is installing *-devel.i386 packages on x86_64 really a sane default for Fedora? I really doubt it."
Apparently, you're wrong. Building stuff for the secondary arch does "Just Work(tm)", allegedly. I relay this 'information' despite the fact that my own experience is the same as yours -- installing *-devel.ppc64 on ppc64 (which uses all 32-bit userspace by default, since ppc32 isn't as broken an architecture as i386 was) really doesn't make much sense.
This is just one of a bunch of problems with the way we currently handle multilib -- it's time we stopped burying our head in the sand and whining that "multilib hurts", and started actually fixing some of the stupidities which make it hurt. It really doesn't have to be as painful as we have made it for ourselves.