Use better the CPUs in the runners to reduce timeouts and total runtime, and parallelize to more jobs.