Potentially huge performance difference.

For the 99% use case, the performance hit fo the ORM is not significant enough to matter. Most projects have many tables, but only one table that actually needs to have any speed optimizations. That one table can go in NoSQL and the rest can be handled by a ORM.

this is a good question. my experience with the django ORM is that in general it resolves any call to an ORM method into SQL and nothing else, but it is not guaranteed to generate the most optimal SQL.

It is however, a somewhat common practice of django devs to to do some post-processing on a queryset in Python. totally acceptable for small querysets with complicated logic, but, yeah, obviously unacceptable for large performance critical queries.

It will only be one query. The django orm won't do any filtering outside the DB that is up to you.

I believe Django would do this as a single query.