Should newer packages from a different arch really advertise themselves as updates to an older package of a different arch?

I've recently discovered a problem where a new version of firefox.i686 can become available via YUM sooner than firefox.x86_64. (long story)

In hosts that pick up updates automatically, they cheerfully remove firefox.x86_64 and install the newer version of firefox.i686, bringing along a ton of other i686 dependent packages and making a real mess to clean up by hand later on.

I realize that I can exclude 'firefox.i686' pretty easily, but my question is more fundamental than resolving my problem - my question is - should a newer version of a package that is a different arch really be considered an update to that package of another arch?

I do not think it should, but I may be missing something so I wanted to check my understanding.

Thanks,
- KSF