I see that Momcat wondered about,

what the common denominator is here

IMO, it seems to attack larger files more than smaller files ... i.e., the longer the file, the more likely it will be inaccurate.  Other than that it seems to be somewhat random. 

Another point ... based on my 6 tests today (downloads), even though the downloaded file content is corrupt, the file size is unchanged in each test ... this suggests that the correct amount of data is being sent (file size, image size, ZIP file size) ... but incorrect values were being substituted at some point. 

This could be quite nasty to track down ... one strong possibility is a bad memory sitck in a router somewhere ... or a bad port, bad fiber splice, bad cross-connect cable ... so many interesting possibilities ... and not all of those possiblle hardware impairments are under the control of renderosity (bondware). 

--
Martin
p.s. As you might guess, I love to troubleshoot/debug ... hardware/software issues ... then again, my idea of fun ... might not be yours.  [grin]