Denormalization can be successfully performed only with thorough knowledge of the application and should be performed only if performance issues indicate that it is needed. One of the things to consider when you denormalize is the amount of effort it will then take to keep your data up-to-date with changes.
This is a good example of the differences between decision support applications, which frequently need summaries of large amounts of data, and transaction processing needs, which perform discrete data modifications. Denormalization usually favors some processing, at a cost to others.
Whatever form of denormalization you choose, it has the potential for data integrity problems which must be carefully documented and addressed in application design.
