A few years ago I was involved in a project that required capturing content from an online document viewer site that used a SWF for each page, and our product basically did a vector-vector conversion from SWF to PDF. One of our competitor's product did the "render each page as an image, and combine the images into a PDF" and the output difference was amazing. IIRC it was around 2 orders of magnitude in size, and 4 in speed of generation. (They used Java, and we used C, which might account for some of that too, so it's not a totally fair comparison.)
I've done similar (though usually less effort) textbook trickery a few times. The Adobe Inept hack is very handy. Oh, and a recent one was stupidly easy: you could view the ebook in your browser, and save excerpts as a pdf, but only 100 pages in total per book. Problem was it stored how many pages you had saved in a cookie, so "Clear the last 5 minutes of browsing history" and you could get another 100 pages, rinse and repeat for all the book and then staple the files together with pdftk.
PDFs typically are full of Postscript (except if they're just scanned images), which is just a text rendering language. As long as you keep the Postscript format valid, you could remove the watermark by just deleting that text.
I didn't know about PDFtk, but Ghostscript can take a PDF and turn it into text Postscript, and it can reverse the process.
But for normal PDFs PDFtk is incredibly useful!