Use multicore-bench with suite order randomization

[polytypic]

This exposes variation due to effects that individual benchmarks have on the runtime state.