What I think is this: if there are people who want to write it, and people who want to read it, no matter the number on either side (as long as it's more than 2 or 3 people total), it's rediculous to send down an iron fence in front of them and say "Andrew Lih and Jiaqing Bao do not believe you should be able to build a separate Wikipedia for this language. Your request is denied."
Similar things to what you said can be said for many minority languages, including for example Sicilian: Formal writing is usually done in Italian, but that doesn't mean Sicilian doesn't exist. Basque speakers often write in Spanish, and Luxembourgish speakers often write in German or French, but that doesn't mean their languages don't exist or that there will be no audience for these Wikipedias.
The problem of audience and readership is a problem for the individual Wikipedias and not the Wikimedia organization, as long as there is the "potential audience" of people who can understand the language when spoken or written (again, there are technologies so illiterates can take advantage of this technology).
You may see zh-min-nan as having "withered", but it's still growing. It would probably help if it had some instructions for the non-initiated on how to read peh-oe-ji, but nonetheless it is still growing, even if it's very slowly.
If some people say "we want a separate Wikipedia for our mother language" and another group says "we don't need a separate Wikipedia, let's use the old one" but the first group says again "we still want this separate Wikipedia, the current solution is not sufficient!", it is inappropriate for the first group to be quashed by the second simply because of a majority or minority.
We have some separate Wikipedias where we'd undoubtedly be able to get by with a unified Wikipedia, but due to issues of nationalism, and since it's never been done before, we haven't tried. But in the opposite direction we have had a couple of similar issues. The traditional vs simplified issue is thankfully now resolved (with the exception of relatively minor outstanding issues), we have a separate Wikipedia for Nynorsk, etc etc.
You talk about people adapting both ways. We aren't talking about traditional vs simplified anymore.
For some time now, Mandarin has totally dominated other Sinitic varieties and in some places there is the stereotype that a Mandarin speaker expects everybody to speak Mandarin, and if they don't, they must be daft (not that everybody fits this stereotype). Accommodations for Cantonese, Minnan, Hakka, Wu, Minbei, Gan, etc. are rarely made, and when they are it's usually in minor local issues, and nowadays it seems it mostly happens in Hong Kong, Macao, and Taiwan.
If I write a section of an article in colloquial Cantonese, and place it on zh:, what will you do? I think there is a good chance you will either remove it as "nonsense" or change it to baihua saying that you are "fixing it" - if this weren't the case, zh: would be a very muddled Wikipedia compared to what it is today.
As Stirling Newberry noted, there is a fairly recent phenomenon of material emerging in "CWY" (colloquial cantonese). Similar signs have been seen from other Sinitic languages (Haishanghua was published in colloquial Wu an eternity ago, but other than that Wu hasn't exactly had a blossoming separate literature; Hakka is starting to emerge as a separate and acceptable variety in Taiwan), but right now I think that by far the most pronounced "linguistic rebellion" (ie, assertion of linguistic independence) is for Cantonese.
Also, a Cantonese Wikipedia would presumably use hanzi rather than romanization, and thus would probably attract a larger crowd than zh-min-nan does (as Bao noted earlier, the fact that it's written in romanization is a turn-off for a lot of people).
The Cantonese dialect does have unique colorful phrases and a different linguistic culture that manifests itself in Cantopop, film and cartoons. Some would seem foreign to "Mandarin" speakers. It would be great to have these Cantonese-isms captured in some way that could be done in a combined ZH Wikipedia.
There is an emerging Cantonese literature, as Stirling Newberry noted.
Mark