I've been working with Django for about 5 years now and I can say it's a great framework for a "quick win" when your use case matches its original intent - that is, if you're building public websites, with a custom CMS backend, and need the usual stuff like authentication, feeds, comments, etc. You can go a long way by just using and extending built-in apps - and for the 20%-40% left you find some great 3rd party libraries for doing what you need.
From my experience, the problems arise when you begin bending it to fit your particular use case. For instance: Django is naive regarding multi-tenancy, something you usually need when building web apps, so you'll have to hack some parts in it; the coupling of most 3rd party apps with the built-in authentication app can also get in the way; etc. If your use case deviates too much from the "website" model, using Flask (or just going down to WSGI) for the plumbing and then picking best-of-breed libraries (SQLAlchemy, Jinja, etc.) can pay off in the long run as opposed to going with monolithic Django from the start because, down the line, you'll find you had to hack most parts anyway.
From my experience, the problems arise when you begin bending it to fit your particular use case. For instance: Django is naive regarding multi-tenancy, something you usually need when building web apps, so you'll have to hack some parts in it; the coupling of most 3rd party apps with the built-in authentication app can also get in the way; etc. If your use case deviates too much from the "website" model, using Flask (or just going down to WSGI) for the plumbing and then picking best-of-breed libraries (SQLAlchemy, Jinja, etc.) can pay off in the long run as opposed to going with monolithic Django from the start because, down the line, you'll find you had to hack most parts anyway.