Problem: In Pb2 valgrind runs, rpl_parallel_seconds_behind_master is failing.

Analysis: Test script is written under the assumption that while calculating
seconds behind_master(SBM), one of the workers should be waiting on 't2' lock.
It is achieved through executing 'LOCK TABLE t2' directly on slave before
'INSERT INTO t2' query (which was executed on Master) is executed by the
slave worker. But the problem is some times slave worker is completing
the execution of 'INSERT INTO t2' before the test script executes
'LOCK TABLE t2'. Hence the SBM value is becoming zero instead of
the expected 3 seconds and test logic is waiting for SBM to become 
greater than 3 and it is getting timed out.

Fix: Changing the logic in such way that 'LOCK TABLE t2' will always
be executed before slave worker sees 'INSERT INTO t2' query.