I've considered it. If I went that route, I'd probably use zstandard instead of PKWARE .zip (even though python stdlib has the zipfile interface built in). Zst

I've considered it. If I went that route, I'd probably use zstandard instead of PKWARE .zip (even though python stdlib has the zipfile interface built in). Zst is waaay faster, the API is more modern and actually architected from the ground up, supports parallel block processing, it's got rfc8878. It's quite nice. I'd have to still implement the object model on top of the block storage. With low compression levels, you can actually tune things to get faster write times than raw bytes, depending on CPU speed vs IO speed. So that's cool. http://facebook.github.io/zstd/ http://facebook.github.io/zstd/