Whatever Kiwichat does aside, one possible solution is to not make visibility a function of radius, but to do a simple slowly time-varying cluster analysis on the fly. Something like a 2-d K-Means (input just being lat/lon points of users), where you periodically update the number of clusters as a function of the total number of users, or something. The challenge that would arise by doing this, would be the scenario where you have a low population density, where you'd get user-groups that aren't super localized (e.g. if you tried this in Montana...). Might not be an actual problem, though.