How "extreme" are we talking about? Some empirical data would be helpful. I'd also be curious to see data about what % of the characters in all the strings in

How "extreme" are we talking about? Some empirical data would be helpful. I'd also be curious to see data about what % of the characters in all the strings in a typical Java program aren't in 7-bit ASCII. I wouldn't be surprised if 7-bit ASCII characters are a large majority of the total, which makes the case for UTF-16 less compelling.