Well, this task is highly parallelizable. Is it not? Why would there be a central bottleneck that you could not scale out?