Benchmark multiple processes with a database each versus one process with multiple databases

[johnchildren]

I am not sure which of the two performs better when running many tests with "fresh" databases, so it would be interesting to get some numbers on this so we can recommend an approach in our docs.