Gil: ..Uh, wow - so, I'm a little surprised you're taking this to the "this can't possibly ever work" level... To be honest, I thought we'd been working pretty well together, and I thought I'd been making great progress. Honestly, I and you both know this will never replace pixel art - that's not the point - the point is to get as close as we can, and I don't think we're there yet.
The reason I took that route is because I went in with graphicsgale, ready to provide to edits that could help you out and found there's very few pixels there that felt placed. I do a lot of cleanups of processed stuff and that's what it felt like, like I'd had to clean the whole image.
I'm a bit skeptic that you'll ever get rid of most of those, but hey, if you're optimistic, I'm all for it. I personally would cop-out at this point and go for a simpler mechanical style that I can get away with more easily. So, sure, let's see how far we can get first, pretty sure a lot of work can still be done.
So, that latest version is wonderful, I feel like the earlier quality is back again. What's more is that the lady looks pretty good for the first time. Not sure why she's behaving so well right now, but I'd try to hang on to that. The dither pattern shows off that feature really well and I like it.
What I'm wondering right now is if there's a way to cluster stuff more. Basically, I see clusters as a bias to join the surrounding colors. In pseudo-algorithm something like: "The closer a pixel is in color to neighboring pixels, the more likely it is to change its color to those pixels". For example, a single grey pixel in a field of white would probably change color to white. A black pixel wouldn't though, it's too far removed from the surrounding white to feel bias. The result would have to be that areas of one color clump together more. This bias would be introduced before dithering, or it will try to "fix" the dither pattern.