The following sequence of operations:

db.coll.insert( { "_id":"keyvaluethat'slongerthan1024length", "a": 5 } );
db.coll.remove( { "a" : 5 } );

results in primary of a replica set having zero documents, but secondary having one document, since remove on the secondary attempts to remove by "_id".

Now if there is a unique index on "a" and you insert a document on the primary with "a":5 the secondary will now crash with duplicate key error while the primary successfully has only a single value.