If a more elaborate method for decaying older games out of the dataset isn't created, how about something as simple as this? The Ratings are run 3 separate times. Once with the last 90 days of games, once with the last 60 days and once with the last 30 days. Then average everyone's ratings from the three runs. This would give a heavier weight to the most recent games without too much re-work.

I like this idea, the straight cutoff of the current version seems like it would really skew peoples rankings from day to day. However, I think Chas's cut offs might be a bit to early. I would rather see a 60-120-180 kind of thing.