Almost every single time I've had to deal with something being slow in the Python web app I work on, it's been a bad query, or bad ORM usage spawning thousands of additional queries to pull in relations.

I've seen a couple out-of-memory deaths (mostly non-paginated db queries, but compounded with the inefficiency of py2), and occasionally an accidentally quadratic loop (which would be faster in C, but usually can just be made fast in Python with more thoughtful code), but it's almost always the DB—oops, I need to add an index, oops I did a cross join, oops, I'm spawning additional 10 queries for each row returned. The time spent in Python on a slow endpoint is usually negligible.