I think the consideration (not addressed by the article), is that high power highly-multi-core systems have the potential to change the kind of computing experience we have, and harnessing the power of these systems effectively with the imperative tools available right now is really hard compared to using pure fp.