I have an absurdly-similar problem, currently using HeroNet cable though and not any fibre. The failure point speed-wise is extremely similar and the services of failure are the same.
Assessing the issue on my end, it appears to happen regardless-of-anything if wifi is enabled in my router (MikroTik hAP mini), and will happen essentially on a timer if wifi is off. This always has only affected devices operating over ethernet, and particularly-exclusively my production desktop, particularly-much-more only since changing my motherboard to a newer one which has a particularly more cabable onboard ethernet NIC than previous, but only in terms of raw capability and not features.
My own, extremely similar issue, notably acted nearly the same when switching to a PCIe ethernet NIC with significantly more features, with exceptions related to the bandaid solution I found.
I can lengthen this aforementioned timer by a significant amount by reducing the target operating clock of my router as well, and keeping my home cold, which has indicated to me in my case that this issue is related to thermal or power-consumption throttling of my router. Looking through this thread, I've seen buffer bloat be mentioned by two other individuals, one of which dismissing, which would be fitting for throttling of the router to cause.
I would check thermals of your router somehow. If it doesn't have instrumentation to retrieve the temperature from within the LAN, try putting a powerful fan over it and see if it stops happening, or starts to take longer. If this is thermals-related, turning off Wi-Fi on the router may severely help as Wi-Fi radios like to heat up devices a large amount, or reducing the target operating clock.
Typically the switch chip inside of a router will heat up severely by the actual rate of all traffic, meaning fast-tracked traffic would heat up the switch more if this is the point of failure and it may remove this issue to turn off the fasttrack in your router, as much as this might reduce network performance.