The whole thing is layered into multiple levels (go check the OSI Model and its Layers on Wikipedia if you're willing to go down that specific information hole ;)) and the physical layer should mainly be handling packet loss on the connection between those adaptors, transparently to the higher layers that just see that as lower bandwidth than the spec for the adaptors (a spec which is really quite optimistic, IMHO).

Yeah, a cable with a metal sheaf wired to the GND level (i.e. Cat cable) is going to be way better at higher frequencies and at isolation from noise that two twisted copper wires were the network signal is shared with a different "signal" which whilst generally 50/60Hz (depending on country) can have spikes and noise at other frequencies, so it's never going to be the same.

However for example at home right now I can get a reliable 100 Mbit/s over a pair of those adaptors from my router to my PC and the speed limitation is actually (I believe) from my old router not supporting Gigabit Ethernet rather than from the adaptors which are supposed to handle up to 500 Mbit/sec.

That said and as somebody pointed out, it only works well if the plugs you're connecting are on the same electrical network, as transversing coils isn't exactly great for high frequency signals.

It definitely can be finicky. I had a portable ac that completely killed the power line ethernet connection when it ran. And my current house I have it in i use from where my router is to where my main TV is and it is unreliable even without that AC unit. So it definitely depends on the circumstances.

I used ehternof power for many years and although I didn't look at packet loss I never had a problem with it.