You're underestimating the difficulty of scaling Mnesia the way WhatsApp did (one source says "In-memory Mnesia database using about 2TB of RAM sharded across 16 partitions to store about 18 billion records." [1]). The prevailing wisdom in the Erlang community was that Mnesia could not be made to do such things. The only well-known user of Mnesia at anywhere near this scale prior to WhatsApp was Klarna. To achieve this, Klarna employed at least one of the original Mnesia developers who wrote proprietary patches for their system [2]. So while you accuse parent of being reductionist, your views of WhatsApp's engineering challenges also seem reductionist. Erlang/OTP software doesn't write and scale itself ... large-scale distributed systems are decidedly non-trivial no matter what tools you've chosen.
[1] http://highscalability.com/blog/2014/3/31/how-whatsapp-grew-...
[2] https://www.erlang-factory.com/upload/presentations/601/EUC2...