What is wrong with PDF? It is actually my preferred format. It supports annotation, bookmarks, highlighting, and is an open standard. PDF 1.5+ files can be reflowed to fit small screens. What's not to like?
It's not a full standard (like Microsoft's .NET, only a subset of PDF is standardized). Flowable text requires manual intervention (tagging) that most PDF authors don't do, assuming they even put text in the PDF rather than just use images of text (the latter is all too common). Even when you do have proper flowable text, other elements don't flow nearly as well. You can't change font faces on the fly, or margins, or other layout functionality that should be user-controllable.
Epub, on the other hand, is a complete open standard, essentially being a subset of HTML and CSS in a ZIP container. It has its own flaws, such as lack of MathML support (complex equations will generally be represented by images), but for 99.999% of books it's a better solution.
More importantly, PDF vs. EPUB is more about doing layout the "old way" vs. the "new way". The old way is paper-centric, where designers have pixel-perfect control of every piece of the layout, down to kerning of the fonts if they wish. This is great when you know exactly how your content is going to be viewed (for example, it will always be printed on A4 paper). PDF represents magazines, paper flyers, paper books, desktop publishing, etc, or essentially the print world. EPUB, on the other hand, is built from web standards. It's designed on top of markup that was initially created to empower the end user. You get all the standard buzzwords, like separation of content and display, that you would if you were building a web page. For narrative books, it's pretty straightforward to make the swich from PDF (paper) to EPUB (digital). Technical books are where things get difficult, and require a perspective shift. For example, if you were writing a technical book for print on paper, you'd probably have a lot of tables, sidebars, indexes, etc. When you go to convert that to an ebook, you quickly find that EPUB is somewhat limited on first glance. You're dead set on replicating the tables and sidebars and such from your printed book, so you just say, "Screw it, ship the PDF." But that's paper-centric thinking. In the digital world, that sidebar would become a link off to other data. The tables could still be there, of course, but you'll have to rethink where they fit in the flow of the text so that they render well on smaller devices. Indexes are trivial, of course. And there's a ton of other stuff you can do, as newer readers (iBooks, Nook Color) on more capable devices have the ability to embed other media and provide more interactive experiences than what you'd get from a piece of paper or a PDF. It turns out that if you approach the problem from a digital perspective rather than a paper perspective, you end up with something that looks different but still conveys all of the information you intended, and in a better way for digital devices.
To look at it another way, back in the 90s when everybody was just starting to write web pages, a favorite method for graphic designers was to composite a layout in photoshop and then chop that up into multiple images laid out in a table (or worse, use image maps!), just as they would do if they were creating a magazine or flyer layout. Those sites were horrible. They wouldn't scale if you needed to change font sizes to make them readable, they wouldn't flow with the size of your browser ("Best viewed at 1024x768" my ass!), and they eventually broke once the box model was standardized and it turned out that Internet Explorer got it wrong (oops!). You don't see those kinds of sites today, web pages that attempt to replicate the exact look of a paper product, and the reason is obvious -- the web is not paper, and trying to force it into a paper design is painful for everybody. Ebooks are the way, and designers will learn sooner or later that they can't shoehorn their paper designs into an ebook and have it work.