If it turns out to be a timing-related bug it may be that the bug was much less obvious on real hardware.
I once had strange effects on V6 if lib1 and/or lib2 were rebuild by me.
Should be not hard to test.
I guess that my investigation and solution may be helpful to others.
The server has ECC memory and there are no messages in the syslog about ECC failures. The SEGV occurred on two consecutive days when systemd was reloading so for the moment I find it unlikely that memory may be the root cause but looking at a core dump would certainly help. Where is it?
find / -name '*core*'