[Sarmat89]

Originally Posted by Hitch

books that require very, very experienced formatters to create them.

And those are minority if we talk about non-/fiction books.

DocBook

For a book, you need what DocBook don't have, and many features are not needed for books.
Also, you cannot choose an arbitrarily complex format like HTML/CSS or DocBook, as the clients will not be able to implement it fully. You need a clear, compact and comprehensive format that everyone can implement in its entirety.

Possibly because, like me, they don't have the information from their author or publisher?

That is demagogy, again. Every book has the cover artist mentioned in its colophon, along with other copyright holders. The problem is that you need a dedicated person in the library to extract it and put into custom metadata. It would be an obvious decision to include this in format itself to enable machine processing.

if you don't put in the name of the cover designer, you can't proceed

You cannot use other people's work without attribution, you know.

Millions--literally--of Amazon publishers have managed to create MOBI files by uploading a Word file. Mobi and ePUB share 98% identical "format DNA."

You need to have some level of structure in your Word file too, to get a hyperlinked TOC required for Kindles and the TOC itself.
Therefore, same methods can be employed to create a semantically structured document.

The XML uses an XSLT to transform it to--wait for it--ePUB. That's how that XML file is subsequently displayed on a device.

That's just one method for simple readers. There are readers that can interpret XML formats directly.

When they're telling their device how to display "all" sidebars, how do they know how small, or how large, sidebars have to be for different books? How do they know how large or how small the heading next to it is intended to be?

Basing on screen dimensions and user font preferences. If you need a sidebar to occupy exactly 43% of screen width and have a 1.113 em margin, you are doing that wrong and that is not the kind of markup which is cross-platform and cross-device.

And these are a mere handful; not particularly complex, not spectacularly simple.

##1 and 3 are not the typical books: there are color photos incompatible with E-Ink screens.
The other have a simple layout: headings, illustrations with captions, unordered lists, inline images, footnotes. Fonts and colors are for user to choose. Also, a reader rendering the semantically-defined footnote could take measures not to break the line before its marker.

You've provided all of us with exactly zero real, concrete examples as to how your idea is "better;"

To make a sidebar, or an illustration caption, with XHTML, you need to implement the HTML layout. Not all readers and devices can do it.
To render a semantic sidebar, you can use any model your reader is working with: you already know what you need to do, you don't need to copy the physical formatting to imitate it.