Question: high CPU usage for a large cluster

[sgulyaevsky]

I'm evaluating a memberlist based discovery on a large cluster (~ 2500 nodes).

The problem is that the prototype continuously consumes multiple (2-3) cores on each node. However, it didn't happen for smaller installations (~100 nodes).
Is this a normal cpu usage for a large cluster or some configuration issue?

I use DefaultWANConfig and ~300 bytes metadata.
Also, an attempt to rejoin initial seeds is performed once a minute. It seems to be the only way to join cluster parts after datacenter connectivity issues long enough to mark another part nodes as failed.