He covered why they didn't want to use a tree like distribution method in his presentation.

It introduces a possible single point of failure for all nodes which are children of a failed node.