It strikes me that posting a tweet and displaying it to another user's account could actually be two separate transactions.

Can't the spam analysis be performed on a just-in-time basis when a client refreshes its @replies?