The benchmark did indeed not demonstrate a quantum speed-up, but it in fairness to D-Wave this was a test designed based on the customers requirements i.e. for them acing this benchmark was good enough to justify investing in this technology.
My understanding is that the algorithm that was comparatively fast on a classical computer was hand optimized by a graduate student, it was not a generic annealing algorithm solver.
But the paper on this effort of 'beating' D-Wave on a classical machine is yet to be published, so this is all from blog hearsay.