The test is failing because the assumption of the test are no longer valid.
The test relies on possibly stale information to decide whether the pipeline is most likely pushed to sbe or not.
Recent changes have introduced 3 mongos for many suites. Once we deploy more then 1 mongos, commands are dispatched randomly against them.
So the router you might pull the information from could be stale.
With 1 router, this can never happen.