It's also valid for HA clustering (corosync/pacemaker). There are some HW vendor implementations (HPE iLO, DELL iDRAC,etc) where the ipmi request will trigger a graceful shutdown (power button press) ,where the system could stuck indefinitely but still accessing the resources (for example Storage), and only then it will go for a ungraceful (which is the expected fencing behavior) poweroff.
. This guarantees that failover is as fast as possible, and also that the same host to be fenced, could create more damages if it reacts and begins the shutdown procedure instead of simply powering off.