It's no surprise, and not even new, that manufacturers are doing things like overclocking their processors to manipulate benchmarks and gain higher scores than they should be getting. What's interesting here with this report is that Huawei doesn't seem to be overclocking their Kirin 960 (the Huawei Mate 9 was in this test, so we're only talking about the Kirin 960 in the Mate 9), but they seem to be freezing it or underclocking it when you perform the same benchmarks multiple times. Now, most reviewers only perform a benchmark once and they are all set, but that doesn't really replicate real-world usage.
Shrout Research released a new report this morning showing the Pixel, Pixel XL, Mate 9 and iPhone 7 Plus in the 3DMark Slingshot test, and showed that the Huawei Mate 9 has about a 12% drop by the fourth run of the test. While the Pixel drops 9% and the Pixel XL at about 4% by the fourth test. The reason for running 3DMark over and over again was to simulate gaming on the device, and also make it an even playing field. This shows that there is definitely something fishy going on here with the Kirin 960 - and possibly other Kirin chipsets. Now whether this is to keep the device cool, or running smoothly when it's put under a heavy load, is unclear as we haven't heard from Huawei yet.
Another test that Shrout Research ran was with GFXBench, and they got about a 34 fps score, and they noticed on GFXBench's website that the median score was 38.1 fps. That's still significantly lower than what they had expected to see, given that the launch of the Kirin 960 last year touted that it could score 51 fps in this same test. Something they did try though was putting their Mate 9 in the freezer and running the benchmark, from there they got a score of 54 fps, which is pretty good and about what Huawei claimed it could get.
It's unclear why the Kirin 960 is being throttled down like this, or really frozen is the more accurate term, but Huawei is not the only vendor that does this. A lot of other vendors do this as well with their chipsets and smartphones. Shrout Research is actually proposing that people revisit smartphones about three months after launch to see how well they still do. Seeing as these scores are dropping over the lifetime of the device - which is expected, especially if you are running battery benchmarks. You can check out their full report at the link below.