Good point. I inherited this design flaw from the snippet he posted on his blog.
mrb, please note that the connections per second and queries per second are handled individually in the benchmark. So the connect handler should make no difference to the queries per second.