Hmmm....
Should a third party intercept or observe the key, the loss of entanglement should be measurable to the recipient since the entanglement is destroyed.
What if the third party uses a system similar to that described in http://www.theregister.co.uk/2013/06/03/quantum_boffins_get_spooky_with_time/ ?
If that could work, they would be able to measure the transmitted photon but still end up with a photon entangled with the original photon which they could resend...