/me facepalms
They open with a spiel about how the x99 system can handle more graphics cards better and how that's an advantage then decide to run test with only one card. By the end of that paragraph they should have had an 'oh, wait' moment, realising that 'making the bottleneck be CPU' isn't always such a good idea regardless of if you're comparing CPUs or GPUs.
Besides anything else, being totally GPU bound and thus the CPUs drawing is also a legit outcome, never mind there is no opportunity for the higher-end stuff to actually flex it's legs.
DigitalFoundry should have thought of this and run a more likely setup for the CPUs - sure, have one set of tests be one card if you really want, but at a higher resolution, also try 2 / 3 / 4 card options as that's where the biggest difference lies. Also lacking minimum framerates is idiotic, it's been highlighted as the most important measurement for a while (though framereate graphs / frametime graphs are better yet to give an idea of if it's a brief dip in one particular loading area or a frequent problem)
You need to get a clue.
The most popular X99 CPU by FAR is the 5820k. It has 28 PCI-E lanes, so in an SLI/Crossfire configuration, it's identical to the 6700k - both are limited to 8x/8x lanes. Only the 5930k/5960X can do 16xX16x.