When LDAP dir-context is used in cluster only EJBs one node are being discovered.
My current analysis of this issue:

• module availability information is initially provided during connection establishment
• intial connection is opened to node1 (the node configured in the client) of the cluster -during this connection discovery code is provided with information about the EJBs on node1
• additional connections are opened during discovery (this results in creation of connection to node2 since the connection to node1 is already established)
• when ldap is dir-context is being configured connection to node2 is never established because before it happens EJB from node1 is already offered to the discovery queue, which results in cancelling all DiscoveryAttempt's which in turn result in closing the connection

Taking the above in the consideration it looks like subtle latency introduced by LDAP can be a culprit here and it seems plausible that actions such as changing on log level may mitigate this effect.