Doesn't seem like their trust model really assumes rational agents.
Of course, people who want to see the pool fail could run malicious nodes, but given appropriate membership restrictions and that fact that all an attack can do is pause the pool temporarily the incentive there should be pretty low.