But note that, as per their benchmark against B-tree's, in memory the cost of the memory fetch is sufficiently small that the cost of the comparisons might wipe out the savings.

When retrieving from disk you have far more cycles to work with before less disk-read sized blocks can get close to competitive.