The thing is, Scrivener doesn’t produce or convert anything to a MultiMarkdown document (at least not to the level you’re expecting). This isn’t at all like the HTML converter—more like a plain-text dump of whatever is in the editor. It has integration for MMD, so that one can write using MMD and produce documents without messing with the command-line, and with some of Scrivener’s bespoke features like Preserve Formatting and compiler headings, we produce relevant syntax, but you’re still supposed to be writing a MultiMarkdown document in order to use the MMD features. That may not always be the case, but that’s how it is right now.
In that case we’d need to wait for Apple to make a rich text to MMD converter, because that is what we are using for the HTML export—and if you dig into threads asking for enhancements to HTML, you’ll find we don’t have much control over that. What tweaks we have made to HTML compile are done with what amounts to search and replace (sound familiar?), which can only solve some problems.
Possible to improve, yes: very difficult to improve, emphatically yes. There is a reason why you do not see a flood of RTF to Markdown-ish converters out there, and of those, you will be hard-pressed to find anything that works reliably and does not require extensive page-by-page proofing.
That’s a good reason to use it, and I’d say we tout that just as much as the rest. It’s just not meant to be a principle that extends between word processing style work and semantic document markups. Think of it this way, Scrivener has two principle ways of working: with rich text or with markup. If you choose the markup route you have all of the formats available to Pandoc (not integrated, but possible with the plain MMD export) and MultiMarkdown integration. If you choose the rich text path then you have everything else in the compile menu available to you.
Like I said before, there is the possibility for a hybrid approach, but it’s definitely more limited, doesn’t work for everything (you’ll never get a highlight turned into CriticMarkup, for instance), and you have to prepare for it with rich text formatting settings that work with the whitespace transformation options (not a problem since you can override formatting with the compiler; it just needs to “be hybrid” in the editor itself).
As an MMD user myself, I don’t really find a compelling reason to use a hybrid approach. I certainly could if I wanted to, but I get a better quality word processor file out of MMD (hierarchical stylesheets, real captions, dynamic cross-references, etc.). Scrivener’s e-book output is visually better than what I can get with Pandoc, but the Pandoc result is easier to style by hand with CSS (which is the more appealing option to me). So I don’t really get the appeal of a hybrid approach—but that’s probably mainly because I prefer writing in plain-text to begin with.