I was also on the front page earlier in the week, and I agree with your numbers. Let's call it 1/s peak. If a page is using CGI, it's often (but not always) be

I was also on the front page earlier in the week, and I agree with your numbers. Let's call it 1/s peak. If a page is using CGI, it's often (but not always) because it's interactive in some way. If users are interacting and generate 10 requests total, on average, you're down to only 100ms per request (which is about how long it takes my system to spin up a new python3 process and `import numpy`).