You can set the value of the karma threshold automatically, to some multiple of the standard deviation of karma for all of HN. So you won't have to constantly revise it.

I assume the distribution of karma here is exponential, so take the distribution's parameter lambda to be the highest value of karma, and set the threshold to half of the inverse of lambda. This page explains it better: http://www.phy.ornl.gov/csep/mc/node18.html