It's the outcome of good architecture, application code, and the fact that single row lookups are already just as fast as a cache, especially in SQL Server running on highend hardware.

By the way, 200M is less than 2500 requests/second which is not much at all. In terms of actual throughput, they are not that big.