I'd say that all the assumptions I have made are far more likely to hold than the assumptions you need to make for quantum key sharing to be safe. That is not to say that I find the quantum key sharing technique highly likely to be broken, but many new discoveries in the field of quantum mechanics are yet to be made.
Why is it impossible to build a computational device that relies on non-observed phenomena, which (as I understand it) are not subject to such limits?
We simply can't tell what limits apply to non-observed phenomena, that doesn't mean there ain't limits, but since we can't observe non-observed phenomena it is quite hard to know their exact nature. That said, it doesn't really matter how many of them there may be, for computation a non-observed phenomena is as good as a TTL gate with no output pins, even if it is doing the right thing on the inside, we have no way of obtaining the generated data.
Quantum key sharing by the way builds on non-observed phenomena, if the so-called non-observed phenomena should somehow prove observable, and thus theoretically useful in computers, quantum key sharing would be broken.
I have already shown humility by saying 512 bits, I highly doubt that 256 bits will ever be breakable, and certainly not within a time frame where encrypted information snatched now would be useful, but the 512 bits argument is a lot easier to lead.
By the way, for anyone not totally into the concept of non-observed phenomena, it is not non-observed as in "We didn't bother to look.", but as in "It is only physically possible to retrieve one piece of data from this dataset, when we retrieve a piece of data the rest is lost, nothing has been influenced by that data, and nothing ever will be, it might as well never have existed.".