It seems unrelated in a way. Flatbuffers and Protobufs are ways to serialize data. The fact that FlatBuffers happen to serialize such that the persistent format is the same as in the in-memory representation is an optimization. It is just as easy to shoot yourself in the foot with one as is with the other in regards to what the article talks about. That is could choose do dump your objects as vertices only instead of including edge information with protobufs, json, flatbuffers, xml, s-expressions etc.
The main point was that serialization needs to be thought about very well, because it will involved compatibility issues. It shouldn't be an automating stream of current object structures to disk.