First you are right about the real reading of the win.
But I think that their first implementation should just
work without much worrying about optional optimizations
at all. Attempts to have multiple things that could be
scalars should be a second pass. Attempts to become
about it (and not promote those ints to full
scalars at the first moment) might take a while.
As for the third point, Larry will have to decide on a
string concatenation operation other than the current.
Personally I can learn a new concatenation operation
fairly easily. As long as he doesn't pick +, I won't
really care what he does there...