The only issues we might face:
1.) This involves the mix-up of user-requirement with the underlying protocol!. No purists would like this.
2.) The spam-generators would start getting intelligent in sending spams - It might start using different 'from' addresses and messages against the same Inbox. If either of the email was delivered, it can safely assume that the email-id is valid. This could be resolved by letting the email-server perform auto-bounce using the same logic it uses currently for spam, rather than explicitly marking it for an 'auto-bounce'.
3.) In cases of the server automatically handling it, we might loose a couple of good emails since it was identified as spam and got auto-bounced.
Can this work ? How else can I avoid that spam from reaching my email box and avoid skimming through the hundreds of spam in search of that one good-email ?