I do think exponential parallelism is a good description of QC, because any adequate causal model of a quantum computation will invoke an exponential number of nodes in the explanation of the computation's output. Even if we can't always take full advantage of the exponential number of calculations being performed, because of the readout problem, it is nonetheless only possible to explain quantum readouts in general by postulating that an exponential number of parallel calculations went on behind the scenes.

Here, of course, "causal model" is to be taken in the technical Pearl sense of the term, a directed acyclic graph of nodes each of whose values can be computed from its parent nodes plus a background factor of uncertainty that is uncorrelated to any other source of uncertainty, etc. I specify this to cut off any attempt to say something like "well, but those other worlds don't exist until you measure them". Any formal causal model that explains the quantum computation's output will need an exponential number of nodes, since those nodes have real, causal effects on the final probability distribution over outputs.