Clustered Zookeeper:
ZK clusters are configured for 2xF+1 for failure. This means that 3 clustered ZK nodes can afford to have 1 node fail. 6 node clusters can afford 2 nodes to fail, etc.
The ZK leader node is the only node which can be written to. All other nodes are read only.
If you have confirmed the previous steps on all nodes, and are still unable to get all components to start up, confirm that all nodes are trying to talk to the leader node. To do this, take a TCP dump of one of the components (ex. MP, PG) while it is starting up and filter on the ZK port numbers (3888 and 2181). Check the IP addresses the traffic is flowing to and confirm its the IP for the Leader node.