I'm liking Andrew's thoughts regarding sensitivity to what spam is, and am thinking about the gmail like vote-if-spam approach.
Wondering if the api community (or really twitterers) would use an api such as this: smellsLikeSpam( list_of_tweet_ids )... Twitter could aggregate and apply policy to resulting votes. If you're doing a twitter interface app, then you've got to provide this unpleasant activity to users which they currentl don't have to do now. But gmail is an argument in favor of it working well and not being too onerous on users in the aggregate. I think the problem in general is more dire for search-based functionality than for general tweeting, since search picks up not only (somewhat) older tweets but new ones and potentially in very large quantities. So if I pick up a tweets yesterday and today it becomes spam, I'll want to know about that and be ale to toss the tweet (god what phrase), which has implications for apps such as mine and for twitter too... .