Amazingly, wikipedia's page
on the subject seems pretty thorough and even handed, and addresses most of what's been brought up so far.
From what I saw looking into things up to finding that bloody page and checking sources thereafter, Gryph's right on a good number of military and police weapons coming from the states (and then nice chunks of them going missing afterwards), Shonus is right that many of the weapons that come into the place aren't auto et al when they cross the border (stuff gets converted afterwards, when it comes to that sort of firearm), and Steeled is right that most of that fancy shit isn't what crosses -- a good chunk of what's used overall, and what is accountable for cartel killings, is handguns and such, not things like AKs. It's still a pretty wide spread of stuff, though. It does seem like a good chunk (if not all) of the fancier US-originating stuff is leftovers from previous conflicts down south (and elsewhere, getting smuggled over), rather than coming across the border.
Actual numbers on what's coming in from the states (either US manufactured or imported and then trafficked) seem to be something no one's quite sure about, unfortunately, and range like freaking crazy. Anywhere from "mere" thousands per year (something like 1-2% of total traffic, apparently) to upwards 80-90% of overall volume (and that's disregarding anything stated by Calderón). Stuff that seems fairly reasonable's point at something in the 20+% range on the low end, though, which is less than I thought. Definitely still pretty damn significant volume, though.
... as a kind of arse covering, though, do note I never claimed that most cartel weapons came from the civilian market, just that at least a plurality likely came from American dealers-- which does seem to at the very least be in question. Significant minority would seem to be a more accurate statement, heh.
----
Couple things I found kinda' interesting:
PDF of a study attempting a statistical estimation of cross-country gun trafficking. Findings are interesting, but the methodology seems somewhat iffy. Don't really know statistics well enough to be able to fully parse what's being said, but there's some interesting stuff in it.
This thing is another PDF of a report done up by some senators. Lot of their suggestions and whatnot can be comfortably skipped over, but there's some interesting stuff in the end-notes and especially the appendix (primarily the response letter from the ATF).