The main difference is that many supposed-to-be-interal links in the "official" generated PDF aren't internal, because the converter seems to have problems with some of them (e.g. when having #hash targets). More importantly, there is no way to colour or underline links with this PDF generator, which kind of sucks because you only recognize links when hovering them.
However, the solution to both problems isn't to go some complicated way to manually generate this manual (pardon the pun), but rather find a good way to create a CHM version of the manual. coda has started on a very simple converter which would simply download all manual pages and images, comment out the navigation code and replace links, but that's not really perfect, mostly because there was obviously a lot of unnecessary MediaWiki markup in it, and clickable screenshots would still point to an external source.
I guess the best way to solve this would be to always embed full-size screenshots and expand them using javascript, but that would require more than simple document downloading and search&replace, and I don't really have the time to fiddle around with it. Optimally, one would use the MediaWiki API for this instead of scraping the website, too.
If anyone else would actually like to give this idea a try, let me know, and I can provide you the already existing scraping code.