First, there are emulators. It's significantly easier to emulate 386 than core2-due i guess, and probably faster. Presence of emulators also means that you have no need to search for old 386 hardware to test 386 support.
Second, Peter seem approach linux kernel like a fully commercial product, he has a 'effort' which may be 'put' into some 'direction' or 'wasted' doing something that do not push 'product' forward. In fact, Linux would be dead if idea behind it dies, idea that software development is interesting. It's usually interesting to run on some unusual hardware, to fix some crappy bug and not 'supporting' such platforms is linux would mean more 'we do not accept patches for this outdated XXX' than 'we spent so many time on supporting it'.
I hope linux would not replicate fate of gnome which in 2.0 branch started from 'removing unusual and rarely used features' and finished up with complete rewrite in new shiny way everybody loves.