That's why so many other posters are calling this the silicon lottery. The differences you're measuring aren't differences between the Samsung vs TSMC processors. They're differences between the individual processors themselves. That is, if you were to test two Samsung or two TSMC iPhones, you'd still see this type of variance in the results.
This is true, and in the article I briefly discuss the concept of processor binning. Two A9 SoCs made by Samsung, for example, running at the same frequency may require different core voltages. This will certainly add some scatter to the data points.
On top of this, though, is the difference in process technologies, which almost assuredly have different voltage-frequency curves and leakage currents. Is a sample size of one sufficient to separate the difference in power consumption due to manufacturing variance from process technology, or even determine if one version is "better" than another? No. But our data, together with good data collected by other testers, can ballpark the difference and assuage any fears that there's huge differences between the two versions.
It would be great if we could get 100 iPhones (or whatever a statistically valid sample would be) to get better data, but we do not have $65,000+ to get that many iPhones
Maybe we could start a Kickstarter campaign.....
- Matt Humrick, Mobile Editor, Tom's Hardware