What PDF offers is a consistent, space-persistent, formatted output. For reading longer works, it actually does matter to me where a passage appears on a page, or within a work. Spatial memory is important that way.
I read a lot of material, in a lot of different formats: paged text (manpages, console-mode browsers), formatted HTML, ePubs, DJVU, image-scanned books.
If I'm reading on a largish (9-10") tablet, PDF in one-page-up format is actually really good. Fills the screen, is almost always suitably readable. Scans of old books (thank you, Internet Archive and Google) in particular are a delight -- there's something about reading a century-plus-old library copy with markings and (hopefully not too much) marginalia, as well as the original typesetting and images.
The main problem I have with fluid formats ultimately is their fluidity. I realise that that's perverse, and that there are times when it's a real benefit, but again, I can't seem to get away from that spatial memory thing.
If I'm extracting content from works, I prefer source (LaTeX, Markdown, DocBook, etc.). Though that's another story.
The ability to spin out formats on demand would be an ideal. I'm looking into ways of doing that.