I remember reading about the Human hearing range in a medical journal. The crux of the discussion centered around the frquency range of a typical human. The test data indicated that the range can be as wide as 20hz to 20khz, with wide variances downwards. But that human hearing is most sensitive in the 200hz to 2khz range.
If this is in fact true, then it would make absolute sense to keep a crossover point out of this range. While there are several well employed crossover designs, all add problems into the equation.