It doesn't. Peak temperature was 97C. Also, slamhk linked to the 2017 review where they only stress test the CPU and GPU at the same time (FurMark+Prime95). The 2016 model reaches similar temps and throttles to the similar speeds under the same test - it doesn't reach higher temps or speeds because the power adapter can't cope with it. Furthermore, synthetic tests are used for such stress tests precisely because it's incredibly unlikely that a CPU will be under such load IRL. Think of it as a worst-case scenario.
Again, considering that you can leave the entire cooling system to the CPU only with an eGPU, the 3.3GHz speed and 97C temps with Prime95 are a more impressive statement of performance than 2.3GHz 90C with the dGPU. In my opinion. It also means that in gaming, for example, the CPU wouldn't throttle at all from it's 3.4GHz max turbo speed for four cores with an eGPU. When those gain Boot Camp support, that is.