In my experience with web applications at non-Googley scale, most performance problems are occurring in places like SQL queries, rather than the raw speed of the web application stack itself.

Another really common mistake is tasks being performed inside the web request that should instead be offloaded to background job processing. If something is going to take more than milliseconds, it should be a job. If you see a PR where someone is bumping up the HTTP timeout in order to keep their slow processing request from being timed out by the webserver, they are doing something inline that should actually be a job.