Louis, I am itching to use phrases like "defying common sense" and "enshrining ignorance" but since that would not be polite I will not do so.
I recognize that second paragraph of yours as standard CoGH dogma that I have heard before from at least three different sources. I consider it, weighing my words carefully, to be fallacious, mechanical and self-defeating.
That piece of dogma seems to have a very low opinion of researchers: they can reproduce faithfully what they see, but they cannot reason intelligently. I'd like to think that someone who does genealogical research has in fact seen quite a few FamilySearch images of original documents and developed a certain degree of nous.
I agree that name variations should be respected, and in this case I would not dream of suggesting that the LNAB should be du Preez. But would it spoil some vast eternal plan if we could just agree on Dupree? The simple rule being, don't use accents unless they are enshrined by custom or indicate pronunciation. Or more simply, don't allow one clerk or one transcriber to invent new ways of accenting a name. By all means distinguish betwen Kotze and Kotzé, though. People are touchy on that one.
I can't agree that UTF-8 specials is the way to go. Must your transcribers now scrutinize a website like FileFormat in search of more and more exact ways of rendering the precise appearance of handwritten text? Suppose someone has the habit of using greek e's. Will we soon have people with LNAB van Rεεnεn? If you are really consistent in this "verbatim", many of your Bassons and Vissers should actually be Baſson or Viſser.
If you look at that whole page, you will notice that almost all the letters u have some kind of swish over them, and that there is no great consistency over exactly what it is. Other old texts, e.g. the Allert van Zijl family bible, are like that too. That is the way that clerks wrote a simple, plain u in those days. It can in no way be compared to an Italian ú or a German ü or an Hungarian ű, all of which serve to indicate actual differences in pronunciation.
A major reason for recognizable, searchable LNABs is that WikiTree gets a chance to suggest a match when you create a new profile. This does not happen when the characters in the name are esoteric symbols from a language not even related to the one spoken by the subject or the researchers. I just tried to create a profile, without saving it of course, for "Lucas Dupree", born 1841, died 1900, and was told No close matches were found. Proceed to step 3.
The slope you are now on will also bring in Dúpree and Dùpree and Dũpree and Düpree, each consisting of one or two individuals. Do you honestly, really want that?