> The thing we _do_ need is to have '16-bit byte' arrays > supported just like
> '8-bit byte arrays', to serve as substrate for the String > implementation.

Isn't that a raw implementation detail?  16-bit byte arrays can be represented internally as half-words of a 32-bit word array or as pairs of 8-bit words.  The interface will obviously hide that, and the user should not care (except for relative performance and size taken) how characters are internally represented.  Either alternative representation will suffice for size, and speed differences should be fairly negligible.  (There already are 16-bit data arrays implemented precisely in this manner).

Or am I missing you point?