This counts multicast exception transmissions -- cases where EIGRP had to fall back from its normal multicast delivery because of an issue like a peer being flagged as unreliable for multicast. A rising count here is a direct signal that EIGRP's flow-control mechanism has detected a neighbor that cannot keep up with multicast delivery and is compensating, which can add latency to route propagation on that segment. On a shared Ethernet segment with five EIGRP neighbors, a spike here right after adding a sixth, slower router would point to that new neighbor as the one triggering exception handling.