You parallelize the load from multiple clients (running on separate hardware). There are some open source projects that facilitate this sort of workload (and the subsequent aggregation of results/stats.)