I'm seven years late, but Statistics Canada's postal code data (for FSAs and, if you use the PCCF, LDUs) are not wrong (generally).
Postal codes in Canada can correspond to disjunct polygons, a bit like Russia proper and its territory of Kaliningrad. Both are Russian territory, but Kaliningrad Oblast (once known as East Prussia) is Russian soil. So goes it with Canada Post's FSAs and LDUs. A given postal code can correspond to multiple, disjunct territories.
In a GIS, this makes the dissolve procedure annoying. You can also end up with centroids that, in theory, do not actually coincide with their corresponding postal code.
I assume that the reason Canada Post does this is that originally there were fewer postal codes, and as particular locations within an existing postal code reached a minimum threshold of addresses, that location received its own postal code. While I have not conducted a comprehensive analyis of Canadian postal code geography, my anecdotal observation is that rural postal codes near suburban sprawl tend to have more fragmented geographies than postal codes in the middle of urban areas.