MongoDB replica sets optimistically apply the log (which in mongodb is called the oplog). In MongoDB, the advantage is not so much that it is time-consuming to apply the state machine changes (though it can be). Rather the advantage is that it allows back-to-back dependent operations to be pipelined and it allows concurrent writes to individual documents from separate clients to be pipelined (later operations can execute based on the state of earlier ones that are still replicating).
MongoDB leverages machinery in the WiredTiger storage engine to preserve snapshots of the database that _are_ durably replicated, to facilitate rollback, but it is still a heavy process compared to an individual write. As such, it is really only useful when elections are infrequent in practice. Throughput definitely drops a lot when elections and rollbacks are frequent, and so in those situations it would probably be wise to apply operations pessimistically.
-Andy