Race condition in starting multi-source replication slaves

Doing start all slaves on an instance that has multi-source replication has produced the following error

130408 4:48:22 [ERROR] Master 'clients': Slave SQL: Unable to use slave's temporary directory /mysql_tmp - Can't create/write to file '/mysql_tmp/SQL_LOAD-' (Errcode: 17 "File exists"), Error_code: 1

Out of the four slaves three of them started, the error was given by the fourth one and slave remains down. I was able to bring it up later with set @@default_master_connection="xxxxx"

It seems all the slave threads from a single instance (multi-source replication) are all trying to use /mysql_tmp/SQL_LOAD-