How about this system:

Let Z = the playing population of the whole pond.

Winner gets + 2Z rating points
2nd place gets +Z rating points
3rd place gets Z/2 (rounded) rating points
4th places gets Z/4 (rounded) rating points

5-8 no rating change

Everyone else loses 3 times the number of players that remains when you fall in.

Instead of counting "wins and losses", you have an aggregate score in your profile.

If you win a pond with 200 players, you get a score of (200 x 1)/200 = 1.0

If you then come in 4th place in a pond of 50 players, your score is now (200 x 1) + (50 x 4)/[200 + 50] = 1.6

It works kind of like a grade point average, only the lower the number, the better.

Someone who wins every pond will have a "1.0"