This seems to be a matter of the above source reporting a variation of the algorithm that is asymptotically slower than the algorithm given in the paper *and* uses asymptotically more memory. Thanks for posting the paper!