No, because once again your assumptions are wrong. A RAM disk is not the same thing as a `brk(2)` block assigned persistently to the process. It's on the other side of a kernel wall, so all accesses have to go through the kernel, just as access to any other type of "disk" does. I'd therefore expect a RAM disk to be considerably slower than `malloc()`. While it's fine to bring assumptions to the table when testing hypotheses, the system isn't going to lie to you about timing. When you benchmark the system in a given way, it tells you the truth about how long that sort of operation takes. If that falsifies your hypothesis, you have to either give up the initial assumption as disproven or you have to dig deeper. Since digging deeper involves investigating how your OS works, it's off-topic here until you can pull that investigation back full-circle to show that SQLite is operating in an inefficient manner. Most likely, you'll find that SQLite is doing the best it can given the portable facilities available. Until then, your results showing that different methods give different speeds is a valuable result: it tells you which method is fastest, so if that's all you wanted to know, you're done.