Backporting a changeset made for 5.0. Comments from there:

  The fix refines the algorithm of generating DROPs for binlog.
  Temp tables with common pseudo_thread_id are clustered into one query.
  Consequently one replication event per pseudo_thread_id is generated.




mysql-test/r/rpl_temporary.result:
  results changed
mysql-test/t/rpl_temporary.test:
  test to generate problematic drop in binlog to feed it to restarting slave
  to see no stop.
sql/sql_base.cc:
  change in drop temprorary tables alg in close_temporary_tables.