Well only Kovid can say for certain, but I can submit a theory for you, and Kovid can shoot me down for talking nonsense.
IIRC before calibre 0.8 calibre did bring back the actual published date of an edition from the website. Unlike your needs however, I would suggest the vast majority of users are storing fiction/non-fiction novels and are more interested that a book was originally written in 1989, than that the particular edition they have was published in 2010. Certainly I fall into that category - I want to read books in the order the author wrote them, not caring in the slightest what edition I have.
However of the metadata sources that calibre pulls from "out of the box" (Amazon, Google and a few others), I don't believe any of them at the time certainly supported a "first published" date (unlike some of the plugins I have written like Goodreads and FantasticFiction). So when Kovid rewrote all the metadata download stuff for 0.8 he made the decision to instead hard-bake it into the calibre code to do a lookup based on ISBN against Worldcat as a source of first published dates.
Unfortunately there are two problems with this. The first is that people who want an edition date instead of first published date (like you in this case) are totally out of luck. The second problem is that Worldcat is an unreliable source, and from my experience gives me some very bizarre dates at times in comparison to what FF/Goodreads display. That is not to say that the latter have perfect data for every book, but right now it is immaterial what dates they have because calibre ignores their results in favour of Worldcat anyway.
You ask what is the point of the option for overwriting published date - the point is it allows you to either get "a" first published date (from Worldcat) or not get one at all and hence not overwrite whatever value you manually put in the column.
Kovid has said he will accept a patch to alter this logic but it isn't a priority for him to make the change himself. I started looking into it a while ago, but I'm afraid the complexity of the code managing the collation makes it a non-trivial change unless you can work out exactly how and when all the magic foo that is in there fires. It fell into the too hard basket and so I continue to manually correct published dates using values from FF (I have the search the internet plugin configured with a keyboard shortcut to open FF for an author). Perhaps one day when my plugins splurge reduces again I will revisit it, or maybe Kovid will take pity on us and fix it while he waits for a build like he did with another "issue" that had been lurking around for years a few weeks ago