Another approach might be to start by subtracting land not available for most human uses for whatever reason, e.g. up on the Eiger, or in a large national park, or on a military reservation (places like Pendleton and Hunter Liggett immediately come to mind in California).
I'd guess those two approaches converge on a similar number, but using the median density is pure statistics, while removing unused land is more of a judgement call.
The median approach does risk minor artifacts based on the divisions used to calculate density (census tracts in the United States), but since we're looking for a median value, and census tracts are pretty small, it should be reasonably precise.
