Disclaimer: I'm kind of phoning it in today. I might update this later

Do I really need to argue this one? It should be obvious on its face. A flat file, stored on disk with appropriate filesystem caching, will always be faster than some overly complicated relation database system.

Look at BerkeleyDB! It's been around since 1994! That sort of longevity isn't an accident. You put the entire database in a single file and when you need to read data from the file, you already know where it is. A simple f.seek(dest); f.read(num_bytes) is all it takes. It doesn't get simpler than that. When you make things overly complex, you just make them slow and useless.