It is possible for a map/reduce job to be interrupted in such a way that the _id does not exist. When a node with this issue is restarted, a message similar to the following will appear in startup warnings:

 WARNING: the collection 'foo.tmp.bar' lacks a unique index on _id. This index is needed for replication to function properly

Since this collection is never going to be replicated (temporary mr collection - SERVER-13981) there should not be a warning about it not being able to replicate properly