But the channel is NOT noise free.
And for it to have any measure of security provided by the quantum nature, it needs to be a dedicated circuit between all host pairs.
When you have a dedicated circuit, you need physical access to perform any attacks. Quantum or regular, it doesn't matter.
Given a dedicated circuit and an attacker with physical access, any unsuccessful MITM attack becomes a successful DOS attack.
Any successful MITM attack will require the private keys of one of the parties. You need both if you want to hear the replies, but you only need Alice's keys to have Alice say "Bob, you're a fag.".
Getting the key is the "hard" part. Just like it's "hard" to get someone's house keys. You can develop various attacks against the algorithm by playing in the noise threshold. It may take you months, years, or longer. It may not. But if you may have physical access to the source and detector, you could exploit all sorts of weaknesses in implementation, take the damned things, or just sit at Alice's terminal. Kind of like walking up to someone's door and trying to pick their lock, get in through a window, etc.
In the end it's all a key sharing problem, regardless of where you source your keys from.
If you have a dedicated circuit network, then you require physical access to launch an attack. Just as you need physical access to attack a dedicated circuit network using copper cable. Quantum stuff doesn't come into play.
And, as always physical access = Win.
The "security" of quantum communications comes not from the quantum nature, but from the fact that you're going to be using dedicated circuits between each host pair. This will never scale to the masses. Quantum communication along a dedicated circuit will pretty much only work in truly local networks and from Obama's to Putin (let's face it, Putin's in charge).