One thought, the original request seemed to be based on wanting minimal differences in file contents for 'small' changes in the database, so an incremental backup could see just small parts of the file changed. Any conversion of the database into 'dump' records would cause ALL of the file after any record that changes length or is added/removed to move and thus be a new difference. My guess is that unless you spend a LOT of effort to make a dump that works to minimize this sort of difference, the raw database may well be a near minimal difference set, as it naturally wants to minimize how much of the file it changes to record differences as disk I/O is the most expensive part of the database access.