[lambda]

And yet, we've had plenty of software patents which have held up in court and caused significant damage to companies which were found to be infringing, and many more which have backed off due to fear of the same.

Mathematical facts aren't patentable. That much has been established. Perhaps a "pure" algorithm isn't patentable; but all software patents start out with something like "a general purpose computer which...", tying it to hardware and making it patentable. Good luck implementing that algorithm without a general purpose computer.

For instance, the i4i patent is a patent purely on an algorithm, and Microsoft lost that case and had to remove the feature from its software: http://en.swpat.org/wiki/I4i_v._Microsoft . I'm not sure how you can square that with algorithms not being patentable.