Originally this seemed to be an orchestration host capacity issue. Turns out it was a replay host issue. The hosts use so much memory that we need swap or more memory to get a clean chicken dance run. The early years require less memory, the later years require much more. Without a cost analysis seems that swap is the way to go for now.