Only affects machines where MAXALIGN > 4, and is a boundary-condition
case even there, but still surprising that it's not been identified
before.  Also reduce tuple chain move give-up messages from WARNING
to DEBUG1, since they are not unexpected conditions.