If you have a single node and SFSB session creation fails it will attempt to retry on another node, logging the exception at debug level (which is not shown be default). If there is only a single node then retry will fail and the user will be presented with a 'no ejb receiver available for handling ...' message instead of the underlying exception.

Our test suite has a lot of random failures with this message, and I think this may be the reason (some other intermittent exception that is hidden by this).