Because if you only have a single observation the mean and the variance are directly related. It's impossible to have "overdispersion" - you can have lack of fit but due to the mathematics involved with only have a single observation you really can't "have too much variation" since you don't really even estimate the variance in the first place.