Hi Luigi,
it depends a lot on your use case. Fetching a big record is more expensive than fetching a small one, so if you access only a few properties very often, but you have many other properties that are never accessed, then having a linked object containing these other properties will give you some advantages.
On the other side, fetching two documents is more expensive than fetching a big one, so if you often need all the properties, having a single object that contains all will give you some advantage.
My advice is to design your domain model in a clear way (it will make things easier at development stage) and to do some tuning based on the use case
Luigi