When a node decides if it's up to date, it compares the highest last applied optime of other nodes to it's own last applied optime: https://github.com/mongodb/mongo/blob/r3.4.2/src/mongo/db/repl/replication_coordinator_impl.cpp#L2781-L2788

We could go into drain mode faster and skip catch up more often if nodes compared the highest last applied optime of other nodes to their own last fetched optime instead.