I don't support Bill's rule of thumb as an always true result. The best answer is to measure the performance of the array. This is because the array performance depends upon the overlap of the radiation patterns of the individual drivers (also whether you are in the near and far field). I'm finding that even for well behaved individual drivers that the array effects are hard to predict without measurements so rule of thumb thinking isn't always valid.

Note that some drivers--most notable are ribbon and planar tweeters--have reduced vertical axis pattern overlap vs. frequency. Hence, little vertical radiation interferences or mutual coupling occurs between the arrayed drivers. Arraying doesn't really impact the bandwidth for this case. Therefore, Bill's rule doesn't apply for this example.

Jim