There would be more time to determine if the message is spam. If enough people mark it as spam and site is blacklisted or taken down, spam would be removed from everyone.
You could push spam blocking to the end user and client, and simply let them not download messages originating from unknown origins (similar to phone + Caller ID), without having to worry about dealing with spam in a centralized manner.
And in the new application: Address == Public Key, Address book == web of trust.
Your mail agent probably doesn't even bother to check or propagate incoming message notifications from addresses you haven't manually added to your address book (100% trust), or addresses within a certain number of hops from someone you manually added to your address book (friends of friend, boss's employees, etc.). And the mail client could further hide and ignore all of the notifications from below a certain user customizable trust score as well.
A very tight feedback loop for generating content that will trick you, and hosting it for just long enough to see if it works.