After a master crashes during binlogging replication event, where the slave already started to copy it to it's relay log, but didn't finish, a STOP SLAVE on such multi-threaded slave takes longer then expected.
In the test case below, it takes over one minute. That long delay may be a problem for failover automation scripts.
The issue does not happen when slave_parallel_workers=0.
Fully reproducible test case available in the upstream report for the same issue: https://bugs.mysql.com/bug.php?id=96400
Likely related is also: https://jira.percona.com/browse/FB8-105