While the author proposes a solution, I don't believe his solution would provide the user with any more value from the pared-down stream. Instead, it would just minimize the deluge of information. Assuming that one is always interested in the most recent message from a user makes a conclusion that most recent = most important. As we all know from experience that isn't the case.

A better solution would systematically rank incoming message types based on prior determinations of importance of related messages (bayesian) and weight can be added to a particular user's messages (positive and negative) to help the cream rise to the top. The addition of collaborative filtering would help those that aren't interested in flagging their own incoming messages (up/down vote) still reap the benefits of the crowd.