I assume by cluster you mean that there are drivers going both vertically and horizontally. Placement of drivers on the horizontal plane reduces horizontal dispersion and leads to comb filtered response. The overall result, to be succinct, sucks. The only time that cluster arraying works well is when the distance across the array is less than 1/2 wavelength; otherwise pathway differentials between the various sources and off-axis listening positions create massive dips and peaks in response. The only time a cluster array is of benefit is when it is engineered to intentionally deliver a very narrow dispersion on both the horizontal and vertical planes, which can be useful in voice only industrial applications but isn't worth much for music.