Is an Intel N100 or N150 a better value than a Raspberry Pi?

Is the N150 really 0.59 W/flop less efficient than the BCM2712, does the N150 just allow you to reach a higher peak performance if you don't care about efficiency, or is the N150 truely already limited to its optimal perf/watt? We can't actually tell because the article only compared a 35.169 Gflops workload to a 62.067 Gflops workload.

I.e. how efficient the boxes running the same amount of work per second, not how efficient are the boxes when you max them out to different maxes. On the surface using Gflops/W feels like it should normalize that until you consider perf/Watt is almost never linear so you're left comparing apples to oranges in the chart.