Scheduler gets connection refused (interface shows "process lost") when trying to start new tasks

In rare system configurations the Job Scheduler fails to start some tasks, while others are started without any problems.
The failed tasks get "connection refused" when connecting to the main process.
When trying to find a new port to offer to a new task, the operating system seems to "lie" about which ports may be bindet. The OS let's the Job Scheduler bind ports that are already used by the Job Scheduler itself although they should be able to be bound at that moment.