I have a question about scaling hydra. We have configured hydra to call a token hook, which itself reaches out to another API. When I test this at scale, I see that in this configuration, with each hydra replica dedicated ~300m cpu and 240M memory, that each hydra instance seems to be able to handle only about 1 RPS (or rather the number of hydra replicas I end up with scales with the RPS). This is despite relatively short response times from the integrated token hook. It feels very much like I've misconfigured something, but it's hard to say what. We are not running in dev mode, and otherwise hydra is performing admirably. Any suggestions?