When I download a PDF or PostScript file from the web, the browser often saves
the file as <something>.cgi (or .pl or .php or whatever), based on the URL of the application, which generated the document. Such document then cannot be displayed by evince, because it thinks that it has "application/x-perl" (or similar) MIME type. I think evince should determine the MIME type of the local file based on the file contents, not extension (using for example the shared-mime-info package). All other viewers/file readers I know of (eog, gthumb, gimp, xmms, mplayer ...)
ignore the file name/extension, and use the file contents instead. I believe evince should do the same.