Poul-Henning Kamp was on a FreeBSD podcast (~15 years ago) and did an approachable walkthrough of the performance considerations he made when he created varnish. It stoked some serious curiosity in me, i was working on a system that heavily cached IO in userspace (and paid LOTS of money to do so - we're talking 100s of nodes) and here he was saying things like (i'm paraphrasing here) it's pretty dumb to cache your data from disk in userspace. One thing led to another and i scored a nice bonus that year. By the end of that year i had learned about (and put to good use) techniques like IRQ pinning and of course understanding the caching tunables in the linux VFS layer. I can't find the podcast to link to but this document is pretty much on the money for what i remember hearing https://varnish-cache.org/docs/trunk/phk/notes.html

Another one is the Postfix codebase by Wietse Venema. It was notable because it's basically had the square root of 0 vulnerabilities despite being written in C and being one of, if not the most popularly deployed MTAs in the world (so basically a constant target for hack attempts - contemporary products like Exchange were basically a laughing stock for vulns in that time period). Anyway, the architecture of that codebase is bordering on beautiful. It's a total goldmine.

Bonus - a popular code base that made my eyes bleed, nginx. I think it's basically been re-written today but the earliest versions of it were horrible to read. It was fast back then but it was like some sick twisted joke of how to write code. This is not to take away from whoever created it, they still created a monumental shockwave when they released nginx, it was far more performant that anything else.