In my mind, a tree is more heavyweight than a buffer. You need multiple buffers to begin to approximate the smoothing you get from KNN and kernel density. I use multiple trees in my surfing for the same reasons, but more on the order of 10 than 100.
