Pool can't reach consensus after stopping part of nodes without restart of all nodes

Steps to Reproduce:
1. Setup pool of 5 node.
2. Send NYM transaction to check that pool works.
3. Stop indy-node on node 1.
4. Send NYM transaction to check that pool still works.
5. Stop indy-node on node 2.
6. Send NYM transaction to check that there is no consensus.
7. Stop indy-node on node 3, send NYM transaction to check that there is no consensus.
8. One by one start services on nodes 1 - 3 (order doesn't matter), send NYM transaction after each node connection.
=> indy-node is running on each node, but transactions are not written.
9. Restart indy-node on each node and send NYM transaction.

Actual Results:
Pool can reach consensus only after restarting of all nodes.

Expected Results:
Pool should reach consensus when n-f nodes are working.

Additional Information:
After stopping of 2 nodes instead of 3, all works correctly.
It doesn't matter, which nodes are stopped. In case of stopping nodes 3 - 5, behavior is the same.