+jgh 2020/12/20: I don't see why; locals should be quick.
+The defer goes back to version 1.62 in 1997. A local being still deliverable
+during a continued run might result from something like a defer during the
+original delivery, eg. in a DB lookup. Unlikely but possible.
+
+To avoid delaying a local when combined with a callout-hold for a remote
+delivery, test continue_sequence rather than continue_transport. */
+
+if (continue_sequence > 1 && addr_local)