I quite liked this article. That said, I think it misses the two big advantages of denormalized (one big table) data -

It’s much easier for people to use, especially non-technical stakeholders who tend to just not understand joins.

It doesn’t require anything from the query optimizer… until you’ve had a complex query randomly go from “runs instantly” to “spins for hours” overnight, you can’t appreciate the value of a simple execution plan.

A bonus benefit is that it scares off the architecture astronauts!