Oh damn - the problem is fixed now.
Here’s the reason for the problem:
- Virtual environment
- On the Host (VSphere) the NTP wasn’t working, so time was a bit off
- Syncing of time between host and guest was enabled
- NTP Client on guest was enabled.
So the correct time on the client was sometimes overwritten with the incorrect time on the host but nearly immediately corrected again, which was nevertheless long enough to fire those bogous timeouts.
Anyway - thx a lot for your help 