Sorry for the late reply.
I have not been able to find the situation, that triggers the problem.
But it definitely is not related to some particular query or code, it
just depends on machine load.
But two days ago, I doubled the amount of RAM in the machine, and the
problem really seems to disappear.
The system has been stable for more than two days since the upgrade was
done. Compared to the crashes that usually occured many times a day, I
am sure the system is OK now.
But it really is very strange that something like low amount of memory
can trigger such problems. The system even did not use swap space ...
I also suppose the "mysql daemon hangs on Suse" problem, that you
discuss here now, in fact is the SAME problem - at least the indications
are the same. The admin of machine involved really should check his
syslog for the "sigpending lied" messages. When the crash occurs, new
processes are not spawned out immediately - it takes about three minutes
until the first one new process is spawned, so maybe the admin did not
noticed that the amount of processes actually slowly grows.