Their conclusion in section 3.1:
No, a faster language is not always the most energy efficient.
That said, they really seem to be pointing out that there are exceptions to this rule, where the languages that solve a problem most quickly are not the ones that use the least energy doing it. If you step back from the detail of individual algorithms, speed and energy efficiency do seem broadly to vary together, as one would expect.