Isn't this confusing latency with throughput? Mainframes are designed for throughput, raw TXNs per second, and a lot of the architecture is designed around batching. Which is great for throughput, but the antithesis of latency.