OK, here's a summary, starting with the simple case of 1v1.
(Rating1 - Rating2) gives us predicted odds of winning. If the ratings match, it's 50%. If Rating1 is 100 points higher, it's about 90% odds.
A total of 40 points are at stake, and they're split according to the odds. So, in an even match, player 1 will win or lose 20. In a 90% favored match, player 1 will win about 4 or lose about 36.
If the point delta is more than 150 points, the game is not rated, and the lower-rated player gets a handicap bonus of 12% per 100 points difference.
In team games, each player's rating is scored as if he won or lost to an opponent with a rating equal to the average of the other team. So the higher-ranked partner wins fewer points than the lower-ranked one. (This discourages 1900 experts from partnering with 1600 smurfs to pad their scores.)
Unfortunately, handicapping kicks in when any one player is 150 points below the highest, and appears to only adjust those players who are 150 below the leader. (Someone 151 points below gets a big boost, someone 149 points below gets no boost.) We're looking into how it works and how we think it ought to work.