A couple of things from Statistical Assessment Service

In Does Salmon Study Stoke Needless Fears?, Trevor Butterworth adds to what I said earlier. He argues "The case against farm-raised salmon is far from clear", referring to this article by Gina Kolata quoting Michael Gallo as saying

PCB's have not been proven to cause human cancer, and industry workers who would be typically exposed to higher levels have not been shown to suffer a higher rate of cancer

and which tells us that he's not funded by the fish or chemical industries. I've tried to find articles by him, but haven't been successful. It also links to Is Farm-Raised Salmon Bad For Your Health After All?

Mark Woodin, PhD, an epidemiologist at Tufts who specializes in occupational hazards, adds that while “farm-raised salmon probably does have higher levels of PCBs than wild” and while those chemicals certainly do present a problem, it’s very difficult to say just how toxic they are. Scientists cannot expose study volunteers to a substance that is suspected of causing harm, he explains, so most studies are limited to workers exposed to PCBs before they were thought to be dangerous. Workers were oftentimes “wearing no protection and sloshing around in the stuff,” he remarks. That diminishes a study’s usability for the public at large. Then, too, because there is so much variation among the people studied—how much they were exposed to and for how long—scientists are left with imprecise data from which it’s hard to draw firm conclusions.

Even if the strict EPA guidelines were known to be the right ones, they are based on the amount of PCBs that are thought to be capable of causing one additional cancer case in 100,000 people over a 70-year lifetime. Compare that, Dr. Woodin says, to "very good, consistent data that show a couple of coldwater fish meals a week can lower your risk of heart disease, which is the biggest killer in developed nations."