The basic problem you're running into is the immaturity of the epub format.
It's very easy for a publisher to produce an epub 3 of a novel if they're already set up to produce epub 2, and the epub 3 version will typically work fine on readers designed for epub 2.
However, there seems to have been very slow progress to date on getting publishers and device manufacturers going on the new and fancy epub 3 features such as mathml. It's probably an economic issue. Publishers can't wave a magic wand over their catalogs and produce mathml for all the equations that appear in all their books; it would be an expensive case-by-case slog for them. Their profit cows are K-12 and college textbooks, and most of those were designed in a large format that is not suitable for handheld devices. Since the publishers have little economic motivation to start selling epub 3 with mathml, the hardware manufactures have little economic motivation to start supporting mathml in their devices. Apple seems to have partial, lousy support on some of their devices. Meanwhile, Amazon shows no interest at all in making math work on their format. I wish I could hold out some hope that this would get fixed sooner rather than later, but, frankly, the experience with mathml in the browser doesn't encourage such a hope. For example, Wikipedia still doesn't do mathml after all these years. Because of all these factors, there basically is not much progress yet in getting good open-source tools for producing epub 3+mathml.
Since epub 3+mathml isn't likely to become good and mature in the near future, it's worth considering holding off completely on putting a lot of work into converting a book into the format.
Having said that, I do have some experience experimenting with doing this. Basically epub is xhtml, so if you can get xhtml+mathml output from your latex, you're not that far from having a working epub 3+mathml book. There are already lots and lots of tools for converting latex to html. (You listed them in your question.)
There is an open-source program called calibre that will convert any valid XHTML 1.1 + CSS 2.1 document to valid epub 2. What I did was to generate xhtml output, translate it to epub 2 using calibre, and then patch the epub 2 to try to make it valid epub 3. (Calibre is not capable of outputting epub 3+mathml according to the spec, and unfortunately the developer seems to have zero enthusiasm for making it do so: http://www.mobileread.com/forums/showpost.php?p=1904668&postcount=7 . ) Both my book and the scripts I wrote for patching are open source, so anyone who wants to tinker with them is welcome to: https://github.com/bcrowell/calculus . From a brief look at Andrew Stacey's page, it looks like the approach he's used is fairly similar.
The best epub 3 output I was able to produce is here: http://www.lightandmatter.com/calc/ . I don't own an iAnything, but I got one of my students to show it to me on his device, and basically it seemed to have worked to the extent that Apple had implemented mathml correctly on the device. (Their implementation at that time, about a year ago, was pretty awful, though. E.g., integrals signs appeared as boxes.)
Testing is a problem. I use the open-source java program epubcheck to check whether my epub output is valid. However, just because epubcheck says it's valid, that doesn't mean it will render correctly on handheld devices. There will probably be a period of a decade or more during which some people's devices can handle epub 3+mathml and other people's won't. Calibre 0.8.66+ can display epub+mathml properly, but it uses mathjax, which is completely different from the implementation of mathml on handheld readers. Calibre does not currently output epub 3 at all, which is why I wrote the scripts to patch its output to make it valid epub 3.