It enhances the throughput (on average everyone waits less) at the price of a higher max latency (some, who posted a request mobilizing a very-recently-out-of-c

It enhances the throughput (on average everyone waits less) at the price of a higher max latency (some, who posted a request mobilizing a very-recently-out-of-cache-index, will wait way, way more...), isn't it? In the real world those worst cases quite often kill such optimization. The (data size / cache size) ratio and queries local (in time) dispersion are key.