If that alternative is really not documented, or is too difficult to find, the docs should be improved. (And, in that case, thanks for the report!) To me, looking at the code, it appears that the memdb specification variations all bottom-out in the common sqlite_open\*() implementation with SQLITE_OPEN_SHAREDCACHE passed in among the flags. (I would be shocked if those alternative methods of getting the shared effect for an in-memory DB did not bottom out to the same implementation. It would represent a much better optimization opportunity than many others I've seen exploited.) I will see to getting this documented (made easier to find.) In the meanwhile, I no reason not to favor the documented approach beyond the possibility of saving a few bytes of data storage.