What matters most is not 3NF itself, but the reasons behind normalization. Its main purpose is to prevent update anomalies, which normalization accomplishes by storing data in a single location. Conversely, with intentional denormalization, this is managed by updating code across multiple places within a single transaction. Both approaches are acceptable.

Normalization is critical for relational databases and SQL, which were invented to allow non-programmer end-users to access data easily. Therefore, the database must ensure consistency even when a user performs a single update. However, when databases are used by programmed code that has been reviewed and tested, you can duplicate data for performance. This is where MongoDB's document model shifts more responsibility to the developer, leading to improved performance.