Agree with above. If you have a 2 node cluster then if there is a failover, you will be sharing 64GB among three instances. If you don't set max/min for the instances, then you'll have them balancing the RAM as best they can.

Realistically, you'd really want a three node cluster so that if there are failovers, you don't overload the remaining node. While 2 instances will be fighting on a failover for resources, that's better than three instances, which could cause all the instances to appear to stop responding with starved processes.