Hi Keith. Let me rephrase my question. Assuming I have a file on ramdisk on a system with plenty of memory, so the file is already in physical memory (maybe a copy to /dev/null was done to ensure it was freshly-accessed). The file (database) is only 1.6 GB in size, and read-only. I now have three ways of getting Sqlite to access that database: 1. Normal disk operations. 2. Pragma mmap so that Sqlite using virtual storage operations instead of file i/o operations. 3. Open a :memory: database and do a "restore from". Forget about the initial overhead of doing (3). Initial overhead is not a problem. It's only the main application execution that I care about. Would you expect (2) and (3) to produce identical runtimes, since they're both going through the exact same xFetch(), both accessing data that is, and always has been, in physical memory? If the answer is "yes, they will be identical" then my next question is "would you expect a disk file overhead, still on (2) to add no extra time than what would be seen in the upfront overhead of (3)?". If the answer is "yes, it will be near-identical", then that's what I thought too, but I was wrong. The long-running job (ie forget initial overhead) is taking twice as long with (2) compared to (3). (1) is 2.5 times longer than (3). I cannot explain this result. Thanks. Paul.