I have since learned that since the oplog is shared by all the collections, and that is what is being read to keep the index up-to-date, a lot of CRUD operations in a different collection could cause the behavior we’re seeing. Since we do have a lot of writes every half-hour to a different collection, this is likely the issue.