Add benchmarks that measure the overhead of the benchmark harness itself. We want the overhead to be minimal. Also, these numbers are useful to judge the quality of real benchmarks. If a real benchmark's result is too close to the harness overhead, then it's likely not measuring enough of useful work.