Bear in mind I am more of an emulator developer than a ROM hacker. So my view here is coming from what might be considered a different world.
If a programmer is so obsessed with this concept of "accuracy" does any of this really matter? Because if someone is that anal about "accuracy" then the time will come when they no longer support ROM images because that's not "accurate", only a legit cartridge is "accurate".
You're taking that to an unreasonable extreme.
An emulated game on the SNES consists of two basic things: the software and the hardware. The ROM contains the software, but cannot possibly contain the hardware because that's a physical thing. So it's the job of the emulator to simulate the hardware.
Simply: SNES emulators simulate SNES hardware.
Good SNES emulators try very hard to simulate the behavior of what that hardware actually does in real life. Quirks and all. When done properly, software will not be able to tell whether or not it's running on real hardware, or on an emulator. The behavior on each will be identical. This is the goal of "accuracy".
Allowing 8MB of ROM when the hardware only allowed 6 creates a visible change in how the software runs. This breaks accuracy.
The bottom line is... if the emulator is accurate, it will run any game you throw at it. Anything that runs on the system, will run on the emulator. This is invaluable
for ROM hacking and homebrew scenes, as the emulator becomes a reasonably trustworthy test platform. What's more, it future-proofs the emulator.
Inaccurate emulators often have to riddle themselves with dozens of game-specific tweaks in order to get games to run properly. Which is basically how NESticle operated way back in the day. And even older versions of SNES9x and ZSNES. And if you remember, way back in the day there were probably hundreds of hacks and homebrews designed to work in NESticle that simply do not work in modern emulators.
Now imagine if you will, a homebrew game which was designed to run on ExLoROM. A game which was actually tested on real hardware -- so it is confirmed that the game is programmed properly. However, the game fails to run on an inaccurate emulator because of the visible difference in how the hardware is simulated.
Clearly this is not the fault of the game -- but is the fault of the emulator. So this is something emulator authors try very hard
to avoid. The inaccurate emu author would then have to work in some kind of game-specific tweak to get this homebrew to work properly. Which also would fail as soon as someone makes a hack for the homebrew.
An accurate emu, on the other hand, will simply always work. It does what the hardware does, so it can't ever be wrong.
Please expand on your concept of what constitutes a "legit" ROM hack, if you don't mind.
If it runs on real hardware, it's legit. If it doesn't, it's bogus.
Repro carts like Powerpak muddy these waters a lot, as they also tend to be emulators. But that's a side-point.
Now... nevermind all of that... and just look at the consequences of this Zelda 3 IQ hack. Anyone that wants to play this hack will have to abandon their emulator of choice, and use some other emulator.
Nevermind the inconvenience of that... and the fact that it resulted in a HUGE upload of questionable legality because he had to bundle the emulators with his hack... but what if someone wanted to play this on an emu for their phone? They can't. What if someone is running MacOS? They can't play this hack either. What if 5 years from now someone is running a version of Windows that is incompatible with these custom emu builds? That person is also SOL.
By going against what SNES hardware dictates, this hack is destroying its playability.
Conversely, if your hack works on real hardware -- if you obey the limitations of the system... you hack will always
work. Every emulator throughout the rest of time will be able to run your hack.