It is not at all clear to me that quantum computers will ever be practical devices for anything other than simulating quantum-mechanical systems, at which I agree they will have an enormous edge over classical devices. Although "simulating" is perhaps the wrong word: "mirroring" or "modeling" would be better. As Scott Aaronson rightly says, they will be useful for broader purposes only to the extent that the interference properties can be harnessed for our use, as in Childs et al.:

Scott Aaronson: Quantum Computing, Capabilities and Limits: "Every quantum algorithm is... trying to choreograph things in such a way that for each wrong answer to your computational problem... the paths... cancel each other out, whereas the paths leading to the right answer should all be ‘in phase’ with each other.... If you can mostly arrange for that to happen, then when you measure the state of your quantum computer, then you will see the right answer with a large probability.... So nature... gives you a very bizarre ‘hammer’....People always want to... [say] the quantum computer just tries all of the possible answers at once. Bt the truth is that if you want to see an advantage, you have to exploit... interference...