Do any of your processes have severity 40 events in the trace logs?
Another possibility I’ve seen in the past happened when the processes were started with the same contents in their processId files, which are stored in the data directory for each process. Is there any chance that this file was copied between the processes in your deployment? If so, I think only one process with each ID can join. You can resolve this by deleting all of these processId files and letting new ones be created. See How are 'contributing_workers' computed?.