Uploading a preliminary patch that seems to fix this problem, derby-4980-1. The error is that
when starting the Derby server for the slave, the same path name is used for the server's working directory as for the master server ("db_master"). This means that the slave server can't open derby.log for writing since it's already opened by the master server. (Does this fact get logged to the console?)
The patch sets the working directory for the slave correctly to the "db_slave" directory.
SInce the fix is done in the replication test ReplicationRun superclass, other replication tests suffer from the same problem.