AbigailII:
1) The system is the one used by the US chess federation. It is in fact Elo-based like Glicko (Glicko being Elo enhanced with a clever coefficient handling).
2) The system is indeed not Glicko because it handles the coefficient differently. Glicko seems a bit better to me for the reasons you say, but USCF is not evil.
3) The system handles a match as one single entity. I think that is fine : accounting for the score in cubed backgammon matches would make no sense, and even in chess, playing for a draw in order to win the match is fine and should never lose rating points.
4) The system handles a match as one single game. Now that is very bad because the longer the match, the higher the probability that the better player wins. That opens the BK system to exploits.

Backgammon players have devised a fix for the point 4), which is called the FIBS rating. It is not perfect, but it does pretty well. Basically, you multiply the rating difference by the square root of the number of points required to win the match.

I explained to Fencer how it could be quite easily implemented here, and he said something like "maybe" (that was long ago).