Also this is apparently benchmarking Postgres and SQLite _with Python on top_. Some of the overhead here might not have much to do with either SQLite or Postgres at all. Just because one uses an ORM doesn't mean the same code is running and especially considering how sluggish Python is, I'm not convinced by any of this either way outside of this being an improvement for the very specific use-case presented. But we don't know _why_ that is.