And the code had to be compiled with the compilers that even in nineties were "old".

I don't think you are right to say "had to be compiled with". It could be compiled with SunOS cc, and the c compilers of other vendors that were crappy and out-of-date. It could also be compiled on the current gcc/egcs etc. that were out. I think it had issues with e.g. the solaris cc because it was a pretty broken and barely used compiler to begin with.

Just glancing through the qmail GitHub repository gave me a headache. Plenty of files have 0 comments (not really sure where you see the clearly stated assumptions). In addition, he seems obsessed with 1 letter variables which offer no inclination of their purpose at a glance.

Here's something I did for myself a long time ago: take some of the fundamental code like alloc.c, alloc_re.c, fmt_str.c and str_*.c, and start commenting it yourself. I found that it was so expressive and concise that after a short time the code is very clear and the comments get in the way.

The only real noise is that no-one uses SunOS cc anymore and compilers will optimize loops (expanding in place, or doing other magic optimizations when appropriate) so loop-unrolling doesn't apparently buy an increase in performance anymore.

And yet he probably wrote code that would be considered objectively better than anything I have ever seen from my large variable name and lots of comments colleagues. Perhaps there is more to it than surface level concerns?