By now we’re indeed pretty sure that there may be some kind of fundamental microcode conflict between the Ryzen 3600/3600X series of CPUs and BIOS versions released after AGESA 1.0.0.3abb/1.0.0.3ab. Rolling back to one of these versions seems to fix the problem for everyone affected, as far as we can tell - it may not always be easy to do but the results do speak for themselves.
Thanks to everyone who shared their data, it is much appreciated. 