And... ugh.
I just ran a tech manual project through HTML5 export and while it works... it follows the sort of approach I dislike and think is the opposite of creating web-optimized documents.
The good news for many will be that it more or less produces a Publish Online document that is not dependent on Adobe hosting. I'm sure that will go over well with the increasing contingent that want to exploit ID's full page layout power and go to the web for it.
The bad news, in my view, is that it creates hard-defined pages, using the PO/doc-reader flicker arrows, that are closer to PDF images than actual web pages. In fact, it's indistinguishable from FXL (fixed-layout) EPUB except that any browser can display them, without need of a dedicated reader or plugin, with all the headaches that brings. Overall, it's much like a dream project of mine, to create e-books based on HTML5/CSS3, but packaged like EPUB (and there are others out there actually working on implementations, the idea to bring the benefits of EPUB with the power and simplicity of a more basic structural language).
And while hard pages will thrill those who for some reason don't want to use PDF, I find them anything but web-friendly, non-responsive, as much of a kluge and sop to poor practice as FXL EPUB.
On a code level, they use the same practice as FXL: instead of the web-like structure of reflowable EPUB (elements presented in simple code, with styles controlling all), this approach wraps every single element (down to each word) in a massive definition string precisely positioning and styling it. Repeat for 500-1000 elements per page. Completely un-modifiable after export, even though I deprecate that practice for EPUB.
I predict HTML5 export will spark little frissons of joy among the general publish-to-web crowd. I am disappointed at such a retro, limited approach and hope the useful (for creating reflowable long-doc export) "Legacy" HTML export isn't removed or deprecated. It is probably dead in the water, though, with no further updates, fixes or tweaks, which is a loss for a small contingent of users who found it valuable.