respdiff is really slow on Bullseyes

respdiff on Bullseyes takes 91 minutes and identifies 1.5 % query disagreements (80 % of that are timeouts). On Buster the test takes 27 minutes and identifies just 0.34 % query disagreements (8 % of that are timeouts).

Locally I am able to identify something similar but instead of timeouts I get 15 % of SERVFAILs from the 100,000 query set. I can make it more stable by setting jobs = 2 instead of the default jobs = 16 in respdiff.cfg. When running dig manually for known working query, I get SERVFAIL in 99 % of attempts as if the query was over some limit. I don't know what might be the difference between Buster and Bullseye (I checked ./configure and sysctl and there are not significant differences). Though, it might an unrelated problem on my system.

This needs to be fixed for the base image being switched to Bullseye.