This is pretty interesting (I skimmed the later half out of fear of spoilers. Sorry). I'm a subtitle and anti-localisation advocate, but this is by and far the best justified and most reasonable thing I've read on the subject.
However, I'd disagree with the point about fansubs, notes and immersion. In a text and still image based medium, it's largely up to the text itself to provide said immersion. As such I can understand where you're coming from, even if it rankles somewhat.
In a moving image (...mostly. This isn't a commentary on animation quality.) and sound based medium like anime, a large part of the immersion and establishing what you refer to as voice comes from the audio and image themselves. Even with no knowledge of Japanese and no subtitles, one can pick up on tones of voice, emotion, verbal ticks etc. The rather feeble grasp of Japanese that many avid follows of the anime scene have will extend that to sentence structure, a few words and puns and understanding cultural jokes if literally translated.
I find that subtitles for moving image mediums are, and should be, a reference - you look to them for the meaning of the dialogue and you get the immersion and voice from the video. A disparity between things (such as localisation of jokes and honorifics, outright changing dialogue, using different names for characters from the audio track etc) moves your focus to considering the subtitles themselves, and makes you question them from then on. Which of course seriously damages the enjoyment of the show. As for translation notes, as long as they're not massively intrusive (Looking at you, ADV's version of Pani Poni Dash) they provide a deeper reference for those who want it, or can be ignored. Obviously screen space is a premium, so things like separate subtitle tracks for this is ideal simply because taking up space can distract from the video.
That being said I'm basically saying that noticeable localisation is immersion-breaking, and that becomes more relevant as someone's proficiency with Japanese increases. By the same token, they then becomes less dependant on the subtitles. As such it's more an issue with obvious, out of character and meaning-changing localisations. Which you're advocating against.
So good for you.