The collections should be locked at the cluster level so no splits/migrates occur during the backup of that collection. This is probably best done with a new distlock-collection (with an implicit/explicit timeout) command issued between each getmore.