This is a scary editorial, and the preceding comments are also scary. Normalization is not some wonderful thought experiment that it can be nice to have, but it's not strictly speaking necessary. It applies a mathematical rigor to the art and science of relational database development.
Normalization through the third normal form is sufficient for most databases, and most sections of any database. It is derelict of any database designer to suggest that normalization to this point is "optional." The normal forms above that address specific cases that the first three normal forms do not adequately address.
Unfortunately, it is an option and not a poor one to suggest it's optional. In cases like mine, normalization dramatically decreases the performance of reading the data. If you were to come in as a database designer and told me that you were going to normalize the 10 TB of data in our data warehouse just because it's what you have always done with other traditional SMP databases, then you're going to be shown the door.
When it comes to normalization and even denormalization, it just depends...