This was raised in the locale-enhancement OpenJDK project mailing list. Here is the excerpt:

---
Unicode has mechanisms to guarantee that program identifiers are stable over versions of Unicode, and defines properties that have that guarantee: XID_Start and XID_Continue (see
http://unicode.org/reports/tr31/).

However, when I look at the documentation for isJavaIdentifierPart,
isCharacteIdentifierPart, etc., it appears that these are defined not in
terms of those properties, but in terms of properties that are *not*
stable over releases.

http://download.oracle.com/javase/6/docs/api/java/lang/Character.html#isJavaIdentifierPart(int)
etc.

That means that a program that was compiled under one release of Java
could fail under a future one, simply because the identifiers break
under the new release.
---

The entire email thread can be found at http://mail.openjdk.java.net/pipermail/locale-enhancement-dev/2010-October/000284.html