So most of the problems I have with using NoSQL type databases for financial data is related to the default settings and examples encouraging a model of programming, that will result in data loss, migration problems, race conditions, duplicate transactions. It's a bit like PHP in my mind, where you can write solid software, but in order to do so, you have to have very good discipline. I would rather choose a more constrained model, that forces you think about these problems and provides some safety out of the box. So in that sense you are right in your observation that my main issue with it is, that you have to mitigate problems, that shouldn't be problems. Then again MongoDB was never designed with those kinds of problems in mind and targets a completely different set of problems, which in itself is fine.
Edit: I'll add a few more thoughts. Financial data in essence is relational. It is also well shardable if scaling is an issue. There are many constraints that need to be validated and it is a natural fit in my mind.