David, if your argument is that Core M isn't well suited for competing in iPad sized designs, then that's pretty obvious. Intel built a design targeted at next-gen PC form factors, which is biased for performance at higher TDP's, but can scale down to 3.5W for thinner designs. At some point, these kinds of design trade-offs will "cross over" and perform worse than designs optimized for lower power, like A8X - but it's up to OEM's to make that trade off. Based on screen size usability and market preference, it matters more to get the 12.5-13.3" sizes right - which according to the graph you provided, are in the range of 4.5-5W TDP.
Inevitably, some OEM's will decide to push the envelope on thinness. I'm not sure how much it matters, since the difference between 8mm and 7mm is subtle to most people. Some of those OEM's will set a design target for 3.5W, meaning they will have to trade off performance. The market will decide if that's acceptable, but I anticipate most Core M designs to use the 4.5W or 6W TDPs, given that performance is one of the primary selling points for PC systems.
I think many of the benchmarks for Broadwell likely showed performance at a 6W TDP. That's still an improvement over Haswell, since most benchmarks run on Haswell used the 11.5W TDP, not the 6W or 4.5W SDP, and many designs targeted at lower power envelopes had similar throttling issues.
But I guess from my perspective, throttling is part of the design, and what the processor is "supposed to do". That's why benchmarking needs to be done at the system level, because raw product scores are meaningless, if the machine is set to a design point that can't reproduce the higher scores.
As far as I can tell, Intel achieved a >2x performance per watt improvement with Core M over Haswell-Y, and that's mostly a testament of the 14nm process, since Broadwell is more of a "tick" architecture. We might expect some additional architectural level improvements with Skylake, but in the meantime, Core M ain't a bad part at all. I'll leave it up to the review sites to determine if in practice it's a knockout or a dud, but so far we haven't seen anything more than pre-production results on what you'd expect to be a fairly sensitive power-performance curve.
