Fixes #63 (closed)
Compared to 86% of time spent in this function in current master@HEAD with 30 concurrent clients and token list of length 5000, with this change only 3% of time is spent here. This is enough so that HTTP processing overhead dominates the profile.