Multiple tests in replication_recovery_test.cpp check for which order the oplog applier applies them. When using a storage engine that does not support document-level locking the ordering is guaranteed because all jobs get scheduled on the same replication writer thread because the '_id' field is not used when dispatching the jobs. See this code:

https://github.com/mongodb/mongo/blob/a8af69397f70903f07b3433c03bfc8b1ab92399f/src/mongo/db/repl/oplog_applier_impl.cpp#L190-L201

We should fix these tests so they don't assume this order.