There are some really odd results in his benchmarks. For example, he has a Dell Venue 11 Pro with a Core-M beating a Surface Pro 4 with a Core-i5 on some video tests. It is unlikely that a Core-i5 loses out to a Core-M for anything.
He has problems with video testing in general. It seems he couldn't figure out how to test 4K video editing on the iPad Pro. For reference, 9to5 Mac got it to work with some bugginess.
It does not surprise me that a Broadwell CoreM can have faster graphics than a Haswell Core i5. There was significantly more die area allotted to graphics in the Broadwell generation. It's also my understanding that significant effort was spent improving performance/watt in low-power scenarios (sorry, don't have a source for this; I heard it at work).
don't forget the graphics hardware and memory buses. these can have enormous impact on the performance of anything that gets rendered - even ignoring GPUs, if it is pure CPU rendering.
i'm not saying he's not gotten something wrong, just that your reasoning is flawed based on my experience working with these things. measurement should always be trusted over 3rd party numbers imo - but only if you make the measurement yourself.
The Core-M has more execution units (24) than the i5 (20), however, is clocked slower -- 1100 mhz vs 950 mhz. So, depending on how the graphics pipeline is being used, it's rather conceivable that certain tests will be faster on the lower speed processor. Especially with how intel's playing with what sort of graphics unit goes with what chip, counterintuitive results can and will happen.
He has problems with video testing in general. It seems he couldn't figure out how to test 4K video editing on the iPad Pro. For reference, 9to5 Mac got it to work with some bugginess.