It's idiotic complaint really. I switched from Go templates to pre-compiled ones because of engineering wankery of personal project where I wanted to have page render take less than 1ms, and with templates it took like 3-4ms. Even with something complex at say 50ms per page you're looking at ~200req/sec/core which is a lot of traffic
As for "working at scale", this post itself is id ~35 million. You could likely fit all of that into ram on a modestly sized server.