The goal of maximum a/v quality makes sense, and is probably the best. At least it shows care for the viewer over consistency with the rules.
Console accuracy is somewhat of a misnomer, not just because the emulators we use are not 100% accurate, but also because the resulting picture (and in some cases sound) just won't look the same as if it were displayed on a TV. I see three options here.
The first is striving for authenticity: 4:3 (or a similar) aspect ratio for pre-PS2 systems, a small touch of blur, scanlines, and an NTSC filter, no antialiasing or anything. We could actually go really really close to authentic look this way, purists would be happy.
The other option is to go with whatever looks and sounds best, evaluating everything on case-by-case basis. If Sonic fans say disabling PSG high quality is better, then it should be disabled, etc.. Our N64 encodes are also treated
much better than they would have looked on the original hardware, but somehow that didn't raise much (if any) of an opposition as it conflicted the accuracy guideline.
The third option, which I believe to be the most balanced and least intrusive, is to agree upon the default measures and change them to fit specific games or cases like SGB to follow the same goal as in option 2.
By far, most developers know the systems they develop for very well, and take its quirks and usage scenarios into account on the design phase. But miscommunication does occur every so often: for instance,
Progear for CPS2 is designed for a 4:3 arcade monitor, and so every gameplay element behaves accordingly, despite the CPS2 resolution being 384x224 (1.71:1 non-square pixel). However, the artists weren't aware of this fact when designing sprites, so they all appear squashed as a result. This, for instance, puts the "choose the best" scenario in a dilemma: either you have asymmetry in graphics, or asymmetry in gameplay.
As such, it is entirely unreasonable to account for every such mistake, so I believe, whatever options we pick, we do have to agree upon the defaults to fall back to. And that would probably require a bit more technical study (as was conducted some time ago by, IIRC, Flygon who proved the Genesis's 320x224 do not become 320x240 on the actual hardware) than what we've been going with.