There are two scenarios in which it may make sense:
- A chunk of optional attributes that may be associated with a primary entity, which makes it a 1:[0-1] relationship. This may also be used to represent fields of a subclass when performing object/relational mapping.
- A performance denormalization, when done as part of physical design. If the additional attributes are rarely needed, they can be shunted off into a separate table that can be joined in if needed. However, this technique is likely not needed if your database can optimize using a covering index or a materialized view to create a physical representation of the frequently-accessed subset of the data.
