Indeed, I also wrote a critic of one of these price-based estimations: http://blog.zorinaq.com/serious-faults-in-beci/
First of all, note that the majority of mining farms operate at very low overheads—PUE 1.05 or less. I start my calculations by taking into account only system-level cooling overhead (chassis fans, Bitfury liquid cooling tech, etc), and not data center-level cooling overhead, because my power and efficiency figures are measured "at the wall".
For my "best guess" figures, in theory I should take into account the PUE by adding 5% to my figures. But the range is pretty large (470-540 MW), not meant to be ±5% accurate, so I don't think it's useful to add 5%.
For my lower and upper bound, I don't need to take into account the PUE. The lower bound, by nature, needs to assume the overhead is zero. And the upper bound is such a worst-case & unrealistic estimate (by assuming miners always deploy the least efficient hardware of their time) that it must surely already overestimate power by at least 5%, which is equivalent to taking the PUE into account.