End-user validation of embedded consensus works fine and is in fact highly decentralized since validation is the responsibility of every single user and not some potentially-centralized set of validators. Any custom_jsons that violate the consensus rules can simply be ignored. It would mean that the user spent RCs to send a custom_json that does nothing at all. In fact this still relies on Steem's own witnesses to determine and finalize the order of those custom_jsons (since changing the order may change validity), which what happens now.

The most used token in the world, Tether, works this way. It is implemented using Omni, which is an encoding layer on top of Bitcoin. Nothing prevents people from submitting invalid Omni data to Bitcoin, since Bitcoin nodes don't even know anything about it, but the Omni nodes (including those run by exchanges) will ignore/reject it.

I don't know that steem-engine works this way, I've been told conflicting stories including by people who I would expect to know what they are talking about, so I find the whole thing quite perplexing. 

Lack of a clear and well-communicated story on how it actually works is perhaps a better reason why it isn't ready for prime time than any clear technical obstacles, but I'm kind of guessing here, for this very reason.