I prefer to use free applications, as much as possible. But technical value is sometimes a strong enough incentive to switch to proprietary software, such as Adobe Acrobat Reader. Of course, I will give examples to support my claims (pictures down there). I may be mistaken, but please feel free to correct me.
Before I start, let me state that I usually use xpdf for daily work: it is fast, it makes things easy, reload is as simple as pressing "r" or changing a page. But sometimes, xpdf does not do what I would like it to do.
The linear gradient problem
The first problem comes from linear gradient rendering. In one file I did generate (this maybe the reason why the bug does appear there and not in many other places), I use a linear gradient from one colour to another. Only acrobat seems to fill the part of the gradient that is not (indeed) the gradient with the determined colors, as should be when the Extend values are set to true. The reference page is page 186 of the PDF 1.7 reference manual. Quoting from this reference: An array of two boolean values specifying whether to extend the shading beyond the starting and ending points of the axis, respectively. Default value:
[ false false ].
The file is on my local disk and quite heavy, but a smaller version exhibiting the bug can be found there.
gs (and therefore gv, ggv) cannot even interpret correctly the pdf instructions.
xpdf, pdfedit, kpdf, all suffer from the same trouble.
Evince makes even funnier things: 
Only acrobat reader does what I think it should do:
Another pain: large maps and other shadings
I do own yet another file (again, generated by me) that shows a related bug (out of gradient colorisation) and also demonstrate the pain that it is to open a large file (with not that many bitmaps, but many vectorial operations). The file is almost unreadable under xpdf at 200% resolution (it's better than it used to be at 100%, but that may be because my desktop computer uses 4 GiB of memory). Same problem with Evince and pdfedit. Kpdf could not even display the thumbnail. Acrobat reader, rendering the file incrementally, has no problems (it's slow, but works) to display it at 1600% zoom. The bug, here, is that apparently (I will not comment further on that) that free readers render the whole area in memory, even though one needs only to see a very small one. The file is small in octet size but meant to be printed on A0 paper (1 m²) and can be found there.
