When I first started using Django in 2009, it was for a University of Bristol portal application (student reps) which extended central Oracle database held data from the student registration system. Performance was surprisingly sluggish however, especially when compared with using Postgres. As it turned out this was due to the nature of the standard Oracle library in Django, which drops and recreates the Oracle connection for each request. Combined with our network of Oracle databases taking, at worst, a few seconds to establish a connection.
I searched around and compared a range of connection pooling solutions, one clearly out performed the others, but it was just added as an attachment to the django issue tracker. So I packaged this up with tests and have refactored it over time.