This has been a constant low level issue from thread to thread. People suggest using a USB dongle to get an extra port on an SBC like the RPi4, and a small number of people will come along and say that they are unreliable. If they were unreliable, given how many people have started using RPi4s, I would expect a flood of people asking for info. But we don't have actual positive data, what we have is what we can infer from the fact that there is no such complaint thread... So instead here's the thread where you can say what kind of dongle you have, what the chipset is, how long you've been using it, and whether you've had any issues with overheating, disconnects, or hardware failure.
For myself I have a UE300 which uses RTL8153 and it's run as my gigabit WAN continuously for 15 months with no hiccups even.
You're talking about the suggestion that they're, quote, "not made for 24/7/365 use and need to rest and cool down once in a while", end quote? Yeah, no. Nonsense. Absolutely nobody makes ethernet dongles to a "must rest once in a while" specification. ($6 knockoff mystery dongles from aliexpress are another matter: but all bets are off with those sorts of things anyway.)
Also, the cooling of onboard USB-connected NICs, functionally identical but for the fact that they can't be unplugged, is not necessarily improved by being on the same board. A typical embedded system doesn't put a heatsink on the NIC anyway, just the CPU/SoC and (rarely) the memory. If the cooling solution includes a fan it might benefit from that; but they don't make these chips on the assumption that that's the case, and in fact they know quite well that most of these NICs won't get any cooling at all. In which case, a dongle is very likely to be in the cooler setting of the two, surrounded by open air rather than adjacent to much more energy-intensive components.
For that matter, the vast majority of onboard Ethernet controllers on Pi 4s aren't even passively cooled, much less actively...what's the failure rate on those? I've yet to see a single report of a failed Ethernet port on a Pi (except for one where it was yanked off the board by an overzealous and slightly intoxicated user trying to remove a tight-fitting cable).
Also, people's ideas about and attitudes to to USB is based on the overhead and CPU load of USB2 and prior, when it was its own, resource-intensive subsystem developed in part to help Intel drive demand for new CPUs. USB3 is a completely different animal: probably best to think of it as a relatively thin layer on top of PCI-e.
Its "backward compatibility" with USB2 is not inherent in it at all, but by decree: USB3 device manufacturers have to build in the distinct USB2 functionality, whether you want it or not.