If the user waits that the transaction is final, he cannot be defrauded.
In your example, you isolate the merchant from the real network and feed him with a fake branch. The merchant will accept your units and add them to his version of the DAG, but since there are no witness-authored units on your branch, it will not move the stability point forward and your double-spent payment will stay unconfirmed for as long as your attack continues. Number of nodes is totally irrelevant, it is the presence of witnesses what makes a branch real.
Great, I hope now SatoNatomato sees why IOTA couldn't use the same method of peer discovery.
I hope you now see how Iota is fail. It doesnt even have Sybil-defense without resorting to slack-channels, despite the PoW employed which Byteball doesnt need, and Iota expected to be used on IoT devices? Okay!
Iota - Just another pump and dump ICO coin to enrich the founders - which is why you like spreading FUD on something which actually works, and which actually can be used on IoT devices.