redis benchmark high variance in side-by-side tests

I ran some side-by-side tests comparing the redis benchmark in v0.20.0 and v0.21.0. While there have been no recent major changes to this benchmark, the change in throughput measurements has varied from less than 0.5x to more than 2x. Other benchmarks did not have such drastic swings. The benchmark or side-by-side tests may be generating unreliable results.