I just tried running mtest on my test Pyra, as you can't get any more lowlevel and we can rule out kernel settings.
And yes, there surely is an issue with the memory, as you can see in the attached logfile (I stopped after 15 iterations).
Even the "FAILURE" has changed to "FABLURE" in the RAM.
I'll run the memtest with the other PCBs I have here as well!