Have you seen the article in the File::Slurp distribution? At it's core, it's doing something very similar to the "read" approach. I'd guess the speed difference you're seeing is due to the option/error/context checking that it's doing, which your hand-coded example is not.