NAKACK has inconsistent internal state after concurrent node startup

Three nodes are started concurrently. The log from the second node to join shows the following (IPs/ports have been replaced)

05:26:00,594 45102 INFO [org.jboss.cache.TreeCache] (main viewAccepted(): [node1:1234|1] [node1:1234, node2:1234]
05:26:00,732 45240 INFO [org.jboss.cache.TreeCache] (main TreeCache local address is node2:1234
05:26:00,852 45360 INFO [org.jboss.cache.TreeCache] (IncomingPacketHandler (channel=Tomcat-DefaultPartition) viewAccepted(): [node1:1234|2] [node1:1234, node2:1234, node3:1234]
05:26:00,861 45369 INFO [org.jboss.cache.TreeCache] (IncomingPacketHandler (channel=Tomcat-DefaultPartition) received the state (size=1024 bytes)

Then many instances of the following logs for more than a day:

WARN [org.jgroups.protocols.pbcast.NAKACK] (IncomingPacketHandler (channel=Tomcat-DefaultPartition) node2:1234] discarded message from non-member node3:1234, my view is [node1:1234|2] [node1:1234, node2:1234, node3:1234]

ERROR [org.jgroups.protocols.pbcast.NAKACK] (Timer-3 sender node3:1234 not found in received_msgs

For these messages to be logged, NAKACK is in an inconsistent internal state. The addresses in "members" does not match "received_msgs".