KISS principle can be applied to everything. Yeah, even data models and the ensuing databases. Actually in my experience this is the most important place to apply it when designing software that stores data.
When you don't need transactions. Oh yeah your "serious" applications can be designed (sometime even better) without them. When the relations you need can be counted on one hand fingers. Yes that means you would have to know the requirements before coding not after :P . When data amounts are medium (half a trillion smallish documents) on day one. When there are 100+ times data reads vs. one write (most data is written once and accessed orders of magnitud more times).
Then you can take MongoDB, install it on your cheapo x86 server and run it with stock configurations with everything working without any previous knowledge of the DBMS (first time we tried it in production). That also means you don't have the config hook in your code to turn on autocommit mode for your Oracle/Postgre database (sic.) .
U guys just widen your horizons, there is never one best tool for the job. Why use scissors when all you need is a knife?
PS: have used Postgres, Mongo, Oracle extensively on multiple big projects with different goals.