I have now excluded  memory, discs, and disccontrollers as
problem sources. (I ran my memorytestprogram for a very long time.)

New data:

It is definitely a temperature dependent problem. "Primes" do not
crash at 15°C, crash shortly at 20°C, and immediately at 22°C,
which is comfortable room temperature.

I recompiled "primes" by simply typing "make" in its directory.
This version of "primes"was 10% larger, and never crashed!
Definitely weird. There is a difference in the executable code
between those prime programs. Is there someone who knows the
difference in compilation, or a way of examining the executable?

(Please bear in mind that I have to freeze down my apartment in order
 to compile programs. Definitely not comfortable, or healthy, and may be
 impossible in this sudden spell of good weather.)

My suggestion is that this is a processor problems, (or adjacent chips,
such as MMU) maybe related to seldom used registers. Any other
suggestions? Compiling a special version of Pmax that abstain from
using a particular register is perhaps a kludgy solution?

Kim0