I agree that the translation code could use work. But it might be premature
to do more than small bug fixes.
I think it would be better to first think through some of the semantic
issues in Maxima evaluation in general -- for example, variable scope in
programming and in mathematical constructs (like 'sum and 'integrate),
arrays, etc. It would be a pity to put effort into increasing the efficiency
of semantics we'll want to change anyway.
-s