Several nodes (less than f) get ahead the rest ones under load

Steps to Reproduce:
1. Install pool of 1.4.494 version with forced view changes each 1800 seconds.
2. Run low load test (10 txn/s) to fill ledger with 500k txns.
3. Upgrade pool to 1.4.496 version (forced view change interval is not changed).
4. Run high load test `while true; do python3 perf.py -c 20 -n 1000 -k nym -g pool_transactions_genesis; done`.

Actual Results:
View Change was not completed on all of nodes.
Domain ledger sizes:
 - Node1: 655059, 
 - Node2 and Node3: 700497, 
 - Node7 700427,
 - The rest 18: 700227.
Pool stopped working.

Expected Results:
Pool should work.

Logs: https://s3-us-west-2.amazonaws.com/qanodelogs/indy-1473