If you denormalise the database where data it maintained you add to the cost of data maintenance tasks and you run the increased risk of being hit by bugs due to trying to maintain data in more than one place.

If you have separation between where you maintain data and where you present data then denormalisation of the latter makes a lot of sense from a performance perspective.

Obviously a denormlised database is going to require more work when a change is required because data changes end up requiring schema changes