I think the same may be true in many cases with NoSQL. I want to try something different, and it's going to perform better for what I'm doing anyway, so why not.
In some experiments at work we got SQL server to do a bulk import of around 1 million rows (with quite a few interspersed reads) in around 5m 30s. We got the same inserts to happen in Mongo in less than a minute. So, we went with Mongo. Do we need that speed? Maybe, maybe not. It sure has been fun to try something new, though.