If you knew ahead of time the fields that the user wanted to load, by all means do a join and load the data ahead of time. This would work great for RESTful APIs... but the issue with GraphQL is that you don't know ahead of time which fields the user will request, so lazy-loading the data provides efficiency without upfront knowledge of what the user will request.
You don’t need to know what they will request, you know what they did request. The dataloader thing is a kludge because you never know where you’ll have to go to get data for a particular part of the tree - maybe it’s another endpoint.
Having said that, I don’t mind the second request model myself. It’s actually pretty clean and when you take the dB roundtrips out of it, the dB will generally do the same work anyway.
