Exactly. He only knows about the advantages of persistence layers, i.e. serializing objects. But he doesn't talk about the problems of serializing pointers (refs), or even graphs (pointers upwards or to siblings).

A relational structure avoids that, and so you avoid this mess. A OO store or Graph DB can handle that but is an order of magnitude slower because it has to serialize pointers, usually by checking against internal hashes or converting to arrays. Such a store is very nice to use for noobs, but once you understand the complexity of writing to such a store, you happily fall back to design a simpler scheme, which is optimized for fast reading and writing.