Ah, I see. That's a bit inconvenient.
If the image is remotely hosted, various browser may block the image. Remote images are sometimes blocked for security, because they're usually tracking images. I don't know how prevalently remote images are blocked, though; I wouldn't expect enough to account for 1/3 of your stats missing.
I've heard that ISPs such as AOL use forced-anonymous proxies to connect people to the internet. I don't know how frequently that cycles, but if there's a significant amount of time between the hit on your site and the hit on the merchant page (and proxy information changes between those) then the IP address could be different.
There could also be something happening on shared connections, with multiple users accessing the pages from the same IP, etc. Some of these things you could probably find out by analyzing your stats, but it'd be a bit tricky.
I could dream up lots of nasty scenarios eating your stats. I think the bottom line, though, is that no web stats (or any stats, really) are going to be 100% accurate. You just have to design your metrics in such a way that they'll work even if they aren't accurate all the time, for whatever reason; dropped packets, rebellious pixels, etc.