...
This happens, because the agent is queried in parallel for each cluster.
Reschedule active checks on all related Check_MK service which is Spread over 0 minutes also has a negative influence. Always spread it over 1 or more Minutes.
This leads to systemd, xinetd or cmk-agent-ctl blocking the connection.
Also check_tcp or host check commands to port 6556/tcp could consume one per source connection.
...