Assuming the contents of the article are true (which is possible), how long before that happens? A few years? A decade?
On the other hand, no one actually knows whether it's possible to build a quantum computer with enough q-bits that will stay coherent for long enough to carry out the calculation: as you add q-bits, noise becomes more and more of a problem. You can add error correction, but that makes keeping all the q-bits coherent harder because now you've got even more of them! At the moment, no-one knows (unless the NSA has built one and isn't telling!) which effect is going to win out as the number of q-bits increases.
(You can't use a smaller QC to simulate a slower version of a larger one, unlike in the non-quantum computing world: if you need to factor a 1024 bit number and you only have a 1000 bit QC you can't do it, as I understand things.)
Note that Quantum Computers don't help as much with symmetric encryption, unless someone comes up with a much better algorithm. They let you effectively halve the key-space, which is a fairly big deal, but you can get back to the same level of difficulty by doubling up the size of your keyspace: a 256-bit AES key provides roughly the protection of a 128-bit key in a quantum computing world. This is different to public-key encryption, where you can double the key size, but if your opponent has a quantum computer with enough q-bits, they can still break your new key in reasonable time.
(aka. http://www.google.com.au/url?sa=t&rct=j&q=&esrc=... or perhaps more easily explained @ http://en.wikipedia.org/wiki/Quantum_error_correction)
If it turns out that the error rate depends (for physical reasons) on the the size of the system, then for some size of QC, adding error correcting q-bits will be counter-productive. Given that (in public) no-one has made a QC with more than a handful of coherent q-bits, no-one really knows where the limits are: it might well be that error-correction lets you build arbitrary sized coherent QCs, or there might be insurmountable physical limits that prevent that from happening. I look forward to people finding out! It's a fascinating new experimental field of physics.
It's conceptually more similar to a GPU which can solve a specific subset of (non-embarrassingly-parallel) problems in polynomial rather than exponential time, with the published number of qubits essentially being the amount of working memory available.
In practice these machines would probably be found optimizing certain steps in machine learning algorithms.
See the comments from D-Wave here: http://dwave.wordpress.com/2011/05/11/learning-to-program-th...
"We do have a factoring algorithm that I’m going to do a series of blog posts on"
I don't know that much about quantum computing, but I believe it may be possible that D-Wave has managed to figure out how to build a useful quantum computer that is nevertheless not general-purpose enough to execute Shor's algorithm for factoring primes.
This upgrade wouldn't need to happen overnight, either. I bet the first quantum computer that can compute sha-256 hashes at all will do so slower than the latest classical ASICs of the time.