As everybody has said, including the original question, this is a terrible idea and don't do it. I recommend some kind of fibre solution, which can be cheaper than you might think. I recently did an economy addition for a charity with the cheapest ethernet switches I could find that had SFP sockets, with ready-made LC-LC multimode 50 m fibres; this used no more equipment, space, or power than copper ethernet, and surprisingly little money. (If it's just the PLC plugging in, then certainly you'd have to accommodate the media converter and its power; conceivably it might be possible to find a module for the PLC directly.) I don't recommend anything radio-based. If you really, really, can't get fibre down the conduit, perhaps even consider powerline communications, where the data goes down the power line instead of a cable alongside, although general-purpose non-domestic ones can be hard to find, and I have never seen one in professional networking and have no experience to say how well it would work.
But your question is about the practical matter of what happens if you do it nonetheless.
If it's not illegal where you are and somehow you convince yourself that it's safe and you don't get fined for building regulations and you can stand the legal exposure and you find yourself doing it against your better judgement or someone else did already did it ...
You don't get failure, you get unreliability.
It is quite likely you'll get an apparently fully working 100baseT link -- ethernet is surprisingly robust, and you have 20 m and 120 VAC with occasional switching -- and might not notice anything immediately or directly. You might even get away with it with perhaps just a few lost packets. In my experience, however, the problems tend to be indirect and often manifest later. Here are two sorry tales from my experience, one current, one old.
I have a client site which is very similar (25 m three-phase armoured cable adjacent to shielded ethernet running 100baseT, PLC running Modbus/TCP for telemetry, occasional SSH for config.) The site is extremely inaccessible. Everything worked, they thought, and left site. It turns out when a once-per-day event happens, half the time a particular device crashes and stays down. It took 20 days to find the pattern. The event is only about a 1 A inductive load going on, and somehow that's disrupting one specific unit through high-quality German DC supplies. It has cost about four months and 30-person days to debug the situation and find an absolutely ridiculous and unsustainable temporary fix by hideous trial and error (replacing crashing unit with an out-of-production one from a different manufacturer). My client very nearly bust its deadline over this, and still doesn't know how to address the problem without a very, very, expensive trip. My client receives daily phone calls from its client on this matter. I suspect they won't get paid until this is resolved.
A subcontractor cut some corners at a company I used to work for and didn't put in enough cable tray in a machine room. A staff member who did know better but just-this-once-can't-hurt laid a lot of ordinary Cat-5 100baseT patch cables next to single-phase 230 VAC lines (just switches and similar current), 5 m here, maybe 10 m there. Everything worked fine, he thought, and finished installation of the machine room. Unwisely, I didn't push the point though we had some very heated exchanges, and allowed it to continue. Two months later, they found the main UPS was kicking in and draining the batteries and losing main server, switches, and routers at random times of day until they went and reset things. Debug loop was about a week on this, took about a month to finally understand what was happening: it was some very unpredictable interaction between noisy lines and UPS control software. As the time was so long after their work, we never got the contractor to make good, and ended up having to pay to put in all the extra traywork, which was then all awkward and had to be done expensively out of hours, with said staff member working nights for about a week recabling. Disastrous for SLAs and credibility and budget.
Don't do it.
