Some time ago I had a hunch that karma followed a powerlaw, so I mapped the leaderboard out in excel - and sure enough it looked exactly like a powerlaw graph.
The curve I fit to the leader board is,
karma = 25343 * rank^(-.465)
Which has an R-squared value of 0.9791.
Unfortunately, this isn't a good enough fit to meaningfully extrapolate very far.
That is assuming pg is included in the count. I haven't looked at the code so I'm not sure if pg is classified as a user or something else.