This release improves CPU, memory and network utilisation in larger clusters.
Instead of broadcasting topology changes to every peer, they are sent to (2log2) neighbours, so in a very sparse cluster the information will take a little longer to reach everyone but it drops the workload on cluster formation from O(N^2) to O(NlogN).
Also we slowed down the rate at which new connections are added: 100 connections will now take 4 seconds.
Full list of changes v0.3...v0.4