Nice hand waving. There have been at least 3 read-only mirrors of PerlMonks. They were each significantly faster than PerlMonks and likely hosted on significantly less hardware.

I don't believe any of them are still being maintained.

Now, surely a large part of the speed / resource difference is related to none of the mirrors ever having taken a significant fraction of the traffic of PerlMonks.

However, the mirrors were substantially simpler. I would expect their page generation to be at least an order of magnitude more efficient. A read-only mirror has little reason to allow you to log in and thus doesn't have to deal with authenticating each request, loading user settings, rendering nodelets, updating user stats, storing nodes in the form that users edit them while displaying them in a different form, etc.

A read-only mirror really acts like a cache, just storing the nodes as they are to be displayed and just serving up nearly static content. Yes, that will always be tons more efficient. I suspect it is typically closer to two or three orders of magnitude more efficient.