Improve the size estimation for the char buffer

[marschall]

On some encodings, especially UTF8, we can make a better size estimation than simply doubling.