For starters, he is impressed that Bush can get through the following simple sentence without stumbling: "To lay out my juvenile-justice plan in a minute and a half is a hard task, but I will try to do so." The bar seems lowered here, but there are other examples. Fallows points to Bush's ability to stay "on message," which is a rhetorically savvy strategy, but hardly the basis for "winning." Again, we're back to definitions, which are not easily dismissed.
Fallows says that neither Bush NOR Kerry has lost a debate in what he calls "the only way that matters: a serious post-debate decline in the polls or an electoral defeat." That's a pretty shaky way to define winning, in my view; it assumes that people watch the debate and then change their mind if the candidate does poorly. (The criteria for how poorly a debater does are subjective and usually partisan when you're talking about presidential politics.) Are post-debate poll results the "dog eat dog" world you're talking about? Fallows makes it clear that Bush "wins" in debates primarily because his handlers set the expectations so low. As long as his opponent doesn't force him into a major gaffe, Bush "wins." (I'm not the only one using the quotation marks; Fallows does it, too.) He quotes Mary Beth Rogers to this effect throughout the article.
Likeability is a key factor. Bush seems more likeable, so most viewers overlook the substance of W's remarks (or lack thereof) and base their judgments on who they'd like to have a beer with.