I think Laplacian pyramids work better because they don't require blurring the masks before merging. Suppose there is a sharp edge with two sides A and B. Suppose side A should come from image IA and side B should come from image IB. With blurring, side B is impacted by side B in IA and IB. This causes the halos. If we don't blur the edges before merging, then side B is only impacted by side B of IB and not side B of IA.

Laplacian pyramid be able smooth the image and reduces the maximum frequency of image features and reduces sharp contrasts and fast changes.