case conversion in sed substitution does not work

[Clint Adams]

> Hmm, actually my test was flawed, and "case 2" is reproducible on both.
> 
> These two cases seem to demonstrate the problem more clearly:
> 
> mizar:[~] echo abc abc abc | sed -e 's/a/\u&/g'
> bc bc bc
> mizar:[~] echo abc abc abc | sed -e 's/a/\ua/g'
> bc bc bc
> 
> This seems to happen even with the C locale.

Yes, it's buggy, and it's not locale-related at all.
I have a quick and simple fix which doesn't seem to break anything.