When you add a new member to a replica set (or need to resync an existing member), it clones the data from the primary. This causes a performance impact. Instead, it would be better if the initial clone was take from a secondary so only the catchup running through the oplog need take place on the primary.