parallel_test by default divides your tests by your number of cores and runs them each in parallel. There's a startup cost, and some contention, so I find that running on 4 cores is a little better than 2x faster than running normally.