That can definitely be true, but note lots of things are latency-, not throughput-bound. In particular, microservice-based systems are likely to hit 99% latency

That can definitely be true, but note lots of things are latency-, not throughput-bound. In particular, microservice-based systems are likely to hit 99% latency of some service on (many/most/all) requests; I'd expect that a similar trick would be helpful there.