First of all, there were already, shockingly enough, casual players in BC. In fact, most of them never even set foot past Karazhan. The difference is, they had access to almost NO content for them. Basically, what did they have? 5-mans (and heroics once the sunwell patch gave them a bit of gear), dailies and PVP. Metaphysical question: what they were paying the sub for? Answer is: not much. On the other hand, where did that money go? To finance raids that almost no one, percentage-wise, would enter. Imagine that we have a country where the taxes from 95+% of the population are almost exclusively used to build palaces for the remaining 5%. This would strike a lot of people wrong on so many levels. Actually, such countries existed (e.g. some quite recent dictatorships in Africa), and no one in his right mind would call it as the right state of things.
Now, a lot of people who are not into VG economics ask me at this point: "But Tom, vanilla and BC were gaining subs, why?". The answer is, because a lot of casual players didn't reach the limits of their game in BC and/or didn't arrive at a state of internal saturation. Simply speaking, the casual players were either leveling (and leveling took a long time back then, especially if you were inexperienced) or didn't arrive at a state of boredom within the game. But they found the game new (because it was) and fascinating (especially Outland - which is still my favorite continent so far, albeit tied with Pandaria). Which is why the model worked at some point. Now, the above is missing a critical piece of information: the sub churn rate. Basically, the fact that the game gained 1M subs is not the only thing we need to know. Because it can mean two very different situations: for example gaining 1.1M subs and losing 0.1M versus gaining 2M and losing 1M. The net gain is the same, but in the latter situation, the customers are leaving much more rapidly, which is not a healthy situation, because the target market is, after all, finite so eventually there will be no "new blood" to compensate for the churn.
We don't have that information, only Blizzard does. Now, at this point we enter the guesswork domain, but based on what the developers said and my experience of high-tech sector, I think my guesses are quite accurate. I assume that, on one hand, by end-BC the churn was increasing, and dangerously so. The model was going to explode and was no longer defensible (not that it was morally defensible before). Basically, they realized that most players were just seeing the first trash packs of Kara (to paraphrase a wow dev) and that something had to be done to make content (into which a lot of money was poured) more evenly spread along the player-base. This process started in LK and finished more or less only with 4.3 and 5.0.
On the other hand, it is pretty safe to assume that Blizzard devs were kicked in the booty by their finance controllers at Sunwell stage for pouring millions into a raid no one ever saw (<1% of pop iirc). So yeah, the devs got smacked on the head (and rightfully so) which made them rethink their content philosophy a bit. It wasn't immediate (the first tier of LK is questionable in that respect) but they eventually got it right by using a multi-tiered difficulty system. Unfortunately, there was the early Cata retardation to mess things up.