Typically, you would have a list of entities stored somewhere, and you'd use quadtree when you need some collision testing between entities for example, by inserting all of the entities and generating a quadtree, and getting the nearest ones that you need for the tests.
But I was wondering, beyond collision testing, the quadtree could be even more helpful if you didn't have a list, but the entities were directly owned by the cells of the quadtree.
For example, when you need to erase an entity, or find a particular entity, you would have to iterate through your entire list. If the quadtree owned the entities, I could pass the position of my entity and it would only seek through entities that are in the same cell.
Instead of iterating through thousands of entities, it would test 'position contains rectangle' a dozens of time, and iterate through the few entities inside the quadtree cell.
Would this even optimize time, and would it be worth it ?