I wonder if you could train a spam classifier to identify & autorespond (or to randomly just drop/bounce/"help"[1] if you want to automate your current workflow :) ) these messages?

Maybe even a simple regex for <theirdomain>.tld would work.

Probably want to have it after a whitelist-known-contacts step to avoid upside-down-ternetting[2] your friends though.

[1] http://pages.cs.wisc.edu/~ballard/bofh/bofhserver.pl

[2] http://www.ex-parrot.com/pete/upside-down-ternet.html