DjangoCon NoSQL Panel
I was on a panel recently at DjangoCon. The topic was NoSQL in Django. You can watch it above if you'd like.
At Urban Airship we went from PostgreSQL as the data store for our most used data to MongoDB. This happened around January/February. For us, it was the quickest path away from spending all of our time fighting fires around our usage on Postgres and towards building new features. Given our set of limitations and our goals, I think it was definitely the right move. Let me be clear, however: Postgres could have done everything we needed it to. We still use it for a lot of our data. However, we didn't have the time, expertise, or money to be able to implement a good and proper postgres setup that would have alleviated our issues. Mongo was extremely quick to get up and running and handled things well until recently. It's starting to show some pains and there are some low level design decisions that impede our moving forward with it (the first point on Ethan Gunderson's recent post is the most relevant here, but isn't the only one).
Joe Stump has some flame bait-y but good thoughts around using NoSQL. As I mentioned above, a lot of it comes down to constraints.
Architecture and infrastructure is a process. It's fluid. It changes over time as your needs grow and evolve. Eventually, you figure out something that works for the long-near-term. That's about all we can ask for.