Cluster size cutoff

We don't have agreement on what the best cutoff is for saying whether two flakes are part of the same cluster. Right now there's some discussion that 4-5 is good. But what's best? WHY??