I always thought that the reason why you wanted to be (at least) one octave up from the - 3 dB point, was that if that point is too close to the slope itself the crossover provides (where the mains are still audible), that you can get the effect of cascading filters.

For example, let's say you pick the -3 dB point as the crossover point. Well, the crossover drops that another 3 dB so you start off -6 dB in the hole already with the mains compared to the sub at - 3 dB (increasing).

I suppose the *steeper* the crossover slope for the mains, the less danger there is in putting the crossover freq closer to the mains - 3 dB point.