Indeed references can be used to build tree structures or at least makes it possible to navigate as if it was a tree. I mentioned this in the article above. This can indeed be a great solution however to gain all the above mentioned advantages of trees, one needs an actual tree structure (which then also means one gets the disadvantages as well). For example while a proper tree structure can serve to also “inherit” properties, the same cannot really be done with a graph/tree (where a graph is of course a super set of a tree structure) build via references.
That being said, some of the advantages of references you mention in your response would of course be compatible with a tree structure as well. Take the Roger Federer example. If one creates a node “Roger Federer” using references one can of course point articles to that node and then one can make all such references query-able. Using tags in this scenario is imho not ideal as likely as a content author I would actually like to be able to point to a specific place to actually describe who this Roger Federer guy is and what he has done. As such pointing to an actual node is imho way superior in this case.
Additionally to me tags can become a problem when rewording an article. Lets say an article originally mentioned Roger Federer, but that sentence is removed. The tags might be overlooked. Now if instead the reference would be set by essentially putting a reference in the actual text, then if I remove the sentence, the reference is removed as well.
But I guess I am just a tag-skeptic 
At any rate, which approach makes the most sense depends on the specific use case. As such contentful offers quite a range of options but the lack of trees means that for me its not useful in many cases, especially when dealing with large data sets.