Using concurrency to estimate server load