I assembled sample data from my ssytem to test this out because I didn't believe it was generalizable. It's the opposite of the suggestion for Oracle 8/9 and sqlserver in my test databases, and DB2 and DB2 UDB rewrite the subquery as a join.
Tables are indexed on constraint columns, indexed on join columns, cust table has ~25 million rows and order table has ~60 million rows.
I don't have a mysql instance up on these servers to test with. I think you're making a generalization of a mysql-specific suggestion.