if data storage and data analysis were wholly independent activities, the logical separation would make a lot of sense, e.g. dumb database is good at data storage, utilities are good at specific analytic functions.

Due to resource bottlenecks these two activities are currently closely coupled. As a result there is no clear cut answer one way or the other.

One aspect currently is a trade off between managing complexity and performance. Of course there are many other factors. Case by case basis.