What do you expect to happen? We don't have enough users giving karma for donation to sustain a linear exchange rate in the [$20, $20000] range. Unless, I suppose, we give up any attempt at fine resolution over the [$1, $500] range.

In practice, what most people are probably doing is picking a threshold (possibly $0) beyond which they give karma for a donation. This could be improved: you could pick a large threshold beyond which you give 1 karma, and give fractional karma (by flipping a biased coin) below that threshold. However, if the large threshold were anywhere close to $20000, and your fractional karma scales linearly, then you would pretty much never give karma to the other donations.

Edit: after doing some simulations, I'm no longer sure the fractional approach is an improvement. It gives interesting graphs, though!

If we knew the Singularity Institute's approximate budget, we could fix this by assuming log-utility in money, but this is complicated.