I would say: definitely yes. I definitely think it is more common for a programmer to expect that if they make a substring, and never reference the original string again, only the substring would remain in memory after GC. It is more intuitive. Without prior knowledge of the implementation, why would you expect a reference to a substring to hold the entire original string in memory? I don't think it is necessarily sloppy code to interpret the method this way when you do not know the underlying implementation.

The description in the Java 7 API for substring is "Returns a new string that is a substring of this string." That does not suggest any sharing with the parent string; it says "new string".