That's not how the turbo works on this AMD mobile chip. There is maximum time that the chip can sustain turbo and that is exactly 300 seconds. As long as the chip is in it's maximum p-state for the duration of the MT test, it will count towards that 300 second turbo time. Prime95 uses more ALUs, caches, and pipelines and thus hits the thermal limit faster and at the same power runs at a lower clock, but this isn't a thermal limit, it's a power limit.
Indeed, we see the 300-second turbo limit not only on Prime95, but also on lighter, real-word tests like on an Agisoft DC workload.
I'm not contesting that the AMD chip in question (as typically configured by notebook OEMs) will not remain in turbo mode for more than 300 seconds at a time. What I'm pointing out is that you cannot assume that all workloads are steady enough over time to produce the simple sustained behavior illustrated with Prime95. This is especially inappropriate when you are referring to another test that directly contradicts this assumption and illustrates variable performance and power consumption, where power does not always stay at 35W even after the initial turbo period is over: https://images.anandtech.com/doci/16446/Power-Agi-5980HS-Per...
If you look closely at the Agisoft test, you will see that it does not exceed 35W until first dropping under 35W, and indeed the average power consumption still does not exceed 35W after the sustained 35W period.
Therefore the assertion that the 5980HS cannot have been exceeding a 35W average load by very much over a 1000-3000 second long test is correct. As you can see after the 300 second initial turbo period the 5980HS is not averaging much than 35W, while the M1 Max is averaging much more than that for multiple sub-tests.
Indeed, we see the 300-second turbo limit not only on Prime95, but also on lighter, real-word tests like on an Agisoft DC workload.