It does address common real scenarios. However, it also causes issues for other common real scenarios. For instance, say you have a CSV parser where you generally only look at a few columns. Under the old system, it wasn't crazy to implement this by, for each field, checking if any unescaping is necessary, and if not (generally the common case) doing a substring. This produces some garbage, of course, but manageable amounts. This approach abruptly gets _far_ slower under Java 7.

I can see why they made the change, but it absolutely did cause problems for real-world use-cases.