Whoops, a (belated) thanks for your reply!
That's a great point, and an annoyance I started to notice as I used it myself... if I have a free evening this week, I'll see if I can apply PG's Bayesian filtering method to Tweets.
The Twitter data hose is an immensely valuable resource, and I hate to see it muddied up with all that spam. I think it's worth working to improve that.
Please let me know if you have thoughts on how I can improve twhittler!