Just deploy each bean as many times as many datasources you have.
Yes, that was my solution too. I created additional jars in the ear. Each new jar has its own META-INF sub-directory with ejb-jar.xml, jboss.xml and jbosscmp-jdbc.xml files. The additional jars use different EJB bean names but are bound to the same entity and interface classes as exist in the original jar (the binding occurs in ejb-jar.xml) The original jar has the actual classes and it's own META-INF descriptors. The original is bound to one datasouce (in jbosscmp-jdbc.xml) and the others are bound to different datasources. It is not necessary to repeat the classes in the other jars. You do need to identify the additional jars in the application.xml.