Most of this is true, but I found that it completely falls flat on its face once you want to do any reasonable HTML output. htlatex is an abomination that should be destroyed in the fiery pits of hell after some long quest with a ring and some hobbits.
Not really - it is more that the various wrappers from tex to html are written in the spirit of the ancient web, rather than a transform to (another) simple, structured markup language. Somewhat odd considering docbook sgml should be a good match for modern html+CSS. I've considered writing something better, but have come to the conclusion that treating tex as output is just more reasonable (ie write in something like rst/markdown etc).