This is by design.
The agent calls the Nagios Plugin via MRPE every 5 minutes, but it has to deliver the output with every call of the agent (i.e. every minute) because otherwise this would create an UNKNOWN for the service check.
That means that the service check itself is run every minute.
You also see the CRIT as soon as the state changes. The maximum number of check attempts (15 in your example) only delay the notification created by the monitoring core. Until the 15 check attempts are reached the service check is in a SOFT state, after that it’s in a HARD state. This can be used as a filter in the views.