After we get performance results from SERVER-83352, we should investigate if we can make the BSON stream reading of additional oplog entries faster. The default behavior is serially threading the stream, constructing an oplog entry, and writing it into the oplog collection.

Some ideas include: can we split up reading the stream and writing the oplog entries into the oplog collection into multiple threads? Can we devise a smarter way of storing the additional oplog entries in S3 so we don't have to read the stream serially?