Know it's less on a concern with fully-shielded and -terminated (Cat6a+?), so this may be a somewhat historical question.
Interested in theory and/or real-world practice and results. Links welcome too!
The balanced signalling used in twisted pair communications is very robust these days, and then you have various protocols on top of that which can handle small errors natively (eg: TCP retransmissions).
It is highly recommended, and often required by code, to keep low voltage and high voltage (high voltage being 120V AC power lines, which really aren't that "high" of a voltage) separated physically. This can be air separation, keep them several inches apart, or conduit separation (don't ever run them in the same conduit). This is mostly a safety issue in the case of insulation breakdown to keep any high voltage leakage from entering the low voltage cable and equipment.
Similarly, things like minimum bend radius, and untwist amount at termination can often be wildly violated with no ill effects. But, it's still best to not temp fate if not needed. Even max cable length can often be exceeded by 20% or more.
It depends, there are types of data cables certified (insulation above 400 V) to be installed in same conduit as mains wires:
If you have a wire carrying a very high voltage but no current at all, you shouldn't get any interference from it. Likewise, you could have a low voltage cable carrying lots of current (though why, I'm not sure), and that would cause interference.
Your point about code and safety is separate, valid, and very sensible though.
OPs point is that if a 230V wire somehow got damaged it could end up touching the outside of an Ethernet cable, and if the Ethernet cable is not rated for that it could cause serious damage to the attached equipment.
You either need to keep Ethernet away from power, or make power 230V+-tolerant.
Last I checked, it looked like it made more sense to just buy pre-terminated, built-to-length cables.
If you're that worried about upgrade-ability, better stick to fiber.
If you're wiring from scratch, sure put the best you can afford, but if you've got cat3 in your walls already, see what runs, your NICs aren't going to look at the label on the insulation. (For better or worse, ethernet speed negotation runs at 1Mbps and the specs don't contemplate testing conditions and reducing speeds until you get to the multi-gig (2.5/5/10g) equipment. Some equipment will drop to 100M in drivers though)
God has not given me any indication on how the future will play out though, I could be wrong. Then again I might be right just because by the time you need more than cat6 there is a new cat 12 that you need. Only time will tell.
10Gtek sells 10gbit multi-mode SFP+ transceivers for ~10->15 USD per, depending on how many you buy at once. I have ten on my LAN and have had no troubles with them. (Now we "just" need to do something about how absurdly expensive 10gbit NICs are...)
(And no, I had never heard of them before I bought a few. I figured that multi-mode SFP+ transceivers have been around for more than a decade, so there's no reason for them to be 50->80+ USD per... purchased from several of the cheapest manufacturers selling on Newegg, and found that 10Gtek's stuff worked fine.)
Hard-earned suggestion: LABEL your fiber cable on both ends with the LENGTH (and id while you are at it). Nothing worse than running the fiber only to have it a few feet too short in the end.
However, it is true that you can't usually expect fiber to survive the "yank really hard on it until it untwists" method of kink removal.
Similarly, other data cables (Infiniband, SAS, etc.) doesn't care either, but they're both short length when used in copper form and they have ample shielding.
In the most recent case, the clock was interfering with a series of motors, causing an antenna pointing array from getting lost and moving erratically. We had had some issues with the motors in the past so assumed it was an issue there and performed 100s of tests to try understand the issue. When we realised it was crosstalk I was quite red in the face.
In spacecraft we talk a lot of precaution to avoid crosstalk when laying out our harness. Shielding is not always an option.
So it seems there would be a fair amount of difference at extremely low separation distances. But we never worked through high frequency EE derivations, which I assume are more average-current- or capacitance-dominated in terms of effects?
In practice it is mostly a solved issue. Applications like Ethernet, HDMI, and professional audio have been using balanced signal pairs for decades and they cancel out most interference. We figured out how to do that in the 1880s.