I still feel somewhat that Mongo is blamed for things that it is not supposed to do. Removing millions of records surely is not a usecase that Mongo is comfortable with. Or if you absolutely need a schema to protect developer doing silly things, then traditional relational database is surely way to go. Mongo is superbly quick to serve read-heavy applications. I did one project with Mongo & Sinatra and the app was able to reach high transaction/second rates compared to the hardware stack it was running on. Of course we faced issues to design trustworthy design for very high scaling. I think it is one culprit with Mongo, sharding and things. But anyway, we as developers and architects, just need to have right tools for the task we try to solve!