With the introduction of 2.6, certain temporary map/reduce collections have incorrectly been replicated to secondary nodes. This adds additional traffic between replica set nodes. Additionally, these collections do not have an _id value in their documents, which causes scanning of collections during replication on the primary and can impact performance.
Large map/reduce jobs with millions of documents can noticeably impact the performance of the server, increase oplog churn and thus network traffic between replica set members.
There is no workaround for replicating inserts to the temporary collections. If the impact to the server increases to intolerable levels, the m/r job should be moved to a dedicated hidden secondary node to mitigate the issue.
MongoDB 2.6.0 and 2.6.1 are affected by this issue.
The fix is included in the 2.6.2 production release.
Documents in temporary *_inc collections are explicitly not replicated. This restores the behavior prior to development version 2.5.5.