Problems with nodes demotion during load test

Case 1:
Steps to Reproduce:

1. Setup the pool of 7 nodes.
2. Write several transactions.
3. Disconnect one of nodes (not primary) using following command in CLI:
send NODE dest=Gw6pDLhcBcoQesN72qfotTgFa7cbuqZpkX3Xo6pLhPhv data={'alias': 'Node1', 'services': []}
4. Send 1000 tnx using load test (based on indy-sdk).
5. Connect disconnected node back:
send NODE dest=Gw6pDLhcBcoQesN72qfotTgFa7cbuqZpkX3Xo6pLhPhv data={'alias': 'Node1', 'services': ['VALIDATOR']}
6. Send 1000 tnx using load test.

Actual Results:
Node which was disconnected stopped processed transactions (it wrote 3027 when the rest ones wrote 3046).

Expected Results:
Node which was disconnected should work same as another ones.

There is a problem to collect extended logs for case 1 because of case 2, so there are only logs with info level for this case: case1_logs.7z

Case 2:
Steps to Reproduce:
1. Setup the pool of 4 nodes.
2. Demote one of nodes.
3. Try to write 1000 tnx using load test.

Actual Results:
Pool is broken after writing 300 tnx. Following messages appear in logs:

2017-12-14 11:12:08,017 | WARNING | replica.py (821) | dispatchThreePhaseMsg | Node3:0 stashing 3 phase message PREPREPARE{'reqIdr': [['Th7MpTaRZVRYnPiabds81Y', 1513249927978175705]], 'viewNo': 1, 'ledgerId': 1, 'instId': 0, 'digest': '41597a8d40cc9ae0505aaa50fceb3fd84d2d81b48a200af9b19461f732420a28', 'ppSeqNo': 301, 'txnRootHash': 'AVHPJs38TuQ3puSHBRrvCYz51fMUratYLYn9UHgkBnBV', 'ppTime': 1513249928, 'stateRootHash': 'FYhgkCWazT92EMzrzfRRwZHqh66tDiHfZgPmk1j1jKPF', 'blsMultiSig': ['RXUoQnE2XWcbke2ekzRyx2aaWmpVUzByNkDzMmHxVNuamw8MdYFgcsFPQqvmBNM1ck5YazZ16Kk4kHT5SJwmQFKMybZF5WYDhXZ1n6FKheSSccT48onVEW5WupBkHymAzUpWokoiN38miTKNP3VXVtkQz6o1oHPCcaWC9Be1oRJWaB', ['Node4', 'Node2', 'Node3'], [1, '3yg9fiR6inj6S9GTJbEpYmCKyynhw9CjBFAs4QzVk5Kb', 'GYrER3ijU71jQeiy1uW8ftfagXatTwPps9wMUVa4ZQyt', '8iA1MpeX34en2UVmMtESoMDEJZ2MdQHN7A7enh2yfdDm', 1513249927]], 'discarded': 1} since ppSeqNo 301 is not between 0 and 300

Expected Results:
Pool should work.

Extended logs for Case 2: case2_extended_part1.7zcase2_extended_part2.7zcase2_extended_part3.7z

Logs for Case 2: Node1.log Node2.logNode3.log Node4.log