So the use-case where parent-child is a better alternative is when you need to do a lot of updating to the individual documents independently, and those sub-documents also need to be treated as distinct on their own. Generally, though, it's always easier to deal with the data if it's denormalized however.