The real question is: what sort of algorithms do we want people to be able to patent? This is an economic question, not a legal one. At what point does the benefit from disallowing free riding out weigh the cost of granting the temporary monopoly? The problem with the law is that it sidesteps this question. Its based on an 19th century notion of invention, whereas the real value of patents is protecting investment into R&D.
One-click shopping? I believe there was lots of effort put into engineering it, but not so much on the R&D side.
Is a patent examiner qualified to implement whatever decision is made on where that line is? I think it's still going to remain a minefield for a very long time.
I think this is one reason the courts have also not had much luck finding any useful & principled theories of invention to apply. The law assumes a kind of classical "eureka!" theory of invention, where there is normal work and then the occasional Invention, which few scholars these days think is how technological advances actually happen. So if the courts were to try to read a scholarly book on either the history or current state of technological advances and try to make use of its concepts (courts sometimes do that kind of thing, if they can find something that will help them make a principled distinction) there would just be a big mismatch in basic assumptions and framework. So they don't, and instead are stuck with trying to make the best of a kind of cobbled-together theory of what constitutes an invention vs. a non-invention. And at this point, absent Congress significantly rewriting patent law, there isn't really much they can do except continue that process of accretion, by doing line-drawing in the gray areas of the existing case law when an opportunity arises.
TBH, I'm not terribly familiar with what can and can't be patented as per business methods. But technically all things that can be computed, can be computed by enough people doing the arithmetic. Computers just make this much faster.
So if I have 300 people...I dunno, collating wedding invitations, and come up with a way of doing it in software, that's not patentable?
However, a clever algorithm for computational imaging is arguable patentable, but anybody who knows matrix math can probably do it manually.
So is it really a discussion over the magnitude of the assist to humans doing the grinding computations by hand and where that line is drawn? Or to your example, some software running on a general purpose CPU can often be implemented as DSPs or some other specialized hardware, it's obviously no longer software I suppose once it's a DSP, even if the method for doing the computation is the same.