My take on it: you get "infinitely parallel" computation, generating "inifinetly parallel" results that are in superposition. Problem is, at the end you can only access your "infinitely parallel" result via a classical measurement, involving a wave function collapse [1].
For some problems, people have found ways to extract useful information via classical measurements, e.g. Shor's algorithm (in theory breaking RSA/DSA/ECDSA/DH/ECDH style public-key algorithms [1]). However in the general case this does not work (so AES and hash algorithms are safe for now).
I mean that's pretty much it. You can't use a quantum computer like a classical computer with a huge (infinite) number of compute cores. Its not a gpu. That's not how it works.
If you use the metaphor that superposition is like computing many things in paralell, the problem comes in that when you measure. The superposition collapses to a single answer at random (with probability related to the amplitude of each possibility) which will usually not be the answer you're interested in.
Explain