Remote Execution fails with SCHEDULER-286 when used too often

When jobs are executed frequently by Remote Execution on a Windows Scheduler, that Scheduler will stop accepting connections after some time and print the message:
2009-07-03 22:22:52.265 [ERROR] SCHEDULER-286 More than 25 communication channels. Connection denied

On the master Scheduler that is controlling the remote execution the following message appears:
ERROR Z-REMOTE-118 Separate process pid=0: No response from new process within 60s

This happens with API jobs.
But it seems to work correctly with shell jobs.