It probably must have been the case that some load runner agent was still ru
nning and calling the application.
However, the infrastructure guys told me that the jvm's that contained the w
eb container (ejb client code) part was allready stopped. Maybe this was reg
istered in the admin console, but the Java process was still active in such 
a state that the nodeagent
could not communicate with it.
We'll have to look more carefully next time.
We haven't seen it happen since.

[ Post a follow-up to this message ]