There are many examples but let's take one: star schemas. I can implement that in MongoDB such that a query can be done in O(1) whereas implementing this in a

There are many examples but let's take one: star schemas. I can implement that in MongoDB such that a query can be done in O(1) whereas implementing this in a relational database requires n joins which in turn requires n foreign key indexes and n row fetches. In many cases this approaches O(n). I have personally seen improvements of 10-100x for this use case when n is large. Also there is a reason that the standard now in enterprise companies are data lakes and not EDWs. Because the cost and effort to model all of your data is prohibitive especially when 99% of the data ends up never being queried.