Once more I would like to thank PMCR for this tip.
Where before I was competely in the dark when getting strange connection problems, this method has now saved me a number of times where I would have been desparate. Also remarkable no tech support has ever mentioned this. Nowadays this on top of my checklist for connection problems!!!
e.g. recently, a project with 7 CJ1Ms on Ethernet to a Dyalox running NS-Runtime.
After a while, my colleague got a call that NS-runtime would occasionally loose connection with 1 host.
After checking all the obvious, I was baffled. In CXP, I could get online with with "auto detect workstation ID" checked with all 6 hosts but not that particular host, but when I unchecked auto-detect for that host it connected.
All in one project tree, using same network settings, same hardware, same routing tables, double checked everything a hundred times. Comms between CPU's with SEND/RECV was working for all hosts.
Then I remembered the above setting. I thought it would make sense if all hosts suffered the same symptoms, but what can it hurt... I changed the adapter to primary, et voila. Instant connection by NS-Runtime and CXP could do auto-detect.
Why only this CPU suffered these symptoms is still unclear to me but at least I was able to solve it!!