In the "Cache-Oblivious Data Structures for
Orthogonal Range Searching" Pankaj K. Agarwal et al. proposed a cache-oblivious kd-tree that its construction is of O(Sort(N)) block transfers. Is there a cache-oblivious version of a kd-tree for orthogonal range searching that its maximum number of block transfers, or cache-oblivious analysis, in construction of it be better than O(Sort(N))=O(N/B logN/B - in base of M/B)?