Off topic to OPs paper however consider the case of volt-db an all in memory SQL db with replication. In the sales'y' write speed was written down to less locks

Off topic to OPs paper however consider the case of volt-db an all in memory SQL db with replication. In the sales'y' write speed was written down to less locks, latches, disk I/O. But somewhere else I read that a big culprit in fact isn't that: it's formatting data to be written to disk then decomposing a disk block back into memory for use. All memory dbs avoid that. Thoughts ?