It does make for a shocking depiction of depths to which cooling can go in relation to slim band of naturally occurring ambient temps.
The reason for failure in these cards has a high probability of being related to unspectacular core component quality the larger electronics world is enduring. It would certainly account for such highly varied behaviors. Modern construction techniques are quite provably robust in terms of environmental (domestic home in Nordic Winter or HK Summer) impact.
On average I'd say component quality has suffered from the supply chain shortages, we do absolutely see increased failure rates, presumably manufacturers have lowered QC to increase output. However we have tested a large batch of cards and their behaviour is identical. They shutdown before they should. We cannot rule out the possibility that adjustments have been made in firmware to account for a lower average chip quality, as in lower the thermal protection threshold, but we don't know this to be true either.
We are currently exploring multiple routes to attain the desired result. As an example we could disable fiber support, due to the additional heat generated by a fiber SFP module a 85C rated controller is downrated to 70C. However I'd prefer to retain the option to use fiber, even though it's currently the "worst sounding" option, simply because of power consumption. To clarify this, the new switch only uses 1.3 watts which includes the power wasted in the voltage regulators, this is a very low figure and a big part of it's performance, 2 SFP fiber modules can easily triple or even quadruple the switch power consumption and multiply noise.
We are also experimenting with heatsinks, interestingly one of the experiments has a surprising large effect on sound quality, I'm currently unsure if it's because of lowered heat or some kind of vibration control effect, but at least there's might be something positive to be gained from this endeavour.