If applications become complex it is, more or less, difficult to ensure a save startup or shutdown of a JBoss node especially within a clustered environment.
Due to this it might be good to start all services and applications before the node join in into a cluster and accept calls or timer-activation.
Also do it vice-versa in case of shutdown.

This shortens the time during the application can no work correctly and will throw Exceptions or not work correctly.

The behaviour becomes more strange if the 'TransactionSticky' features (since 4.2.3 see JBAS-4455) is used, because the UserTransaction service is deployed 'long' before the application-ears are.