Flat Files are the Future
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
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.