Throughput scales with the amount of data in it, it is in the docs. So depending on the application, even if latency is better, the speeds are atrocious at lower volumes of persisted data.