In the float/double question, everybody is making it a real-time vs.
file-time comparison, which I don't think is totally fair.  CSound,
however you use it now, was designed as a file-time program.  In order
for it to be a "real-time" program there are a lot of speed
optimizations possible, some worthwhile (like RTSound) and some
impossibly unworthwhile (like making it run inside a Mac interrupt for
good Mac realtime performance).  The crux of my point is that csound is
a file-time sound generator, which now can be used in real-time for some
things.  Any improvements should be judged first on whether they improve
file generation (e.g. sound quality) and then on how they impact compile
times.  As has been mentioned before, the speed loss from doubles will
likely never be more than 5% and in some case will actually be faster
than floats.
	I'd say it is more worthwhile to worry about RAM issues than speed
issues when considering this switch, since doubles will basically double
the RAM requirements.  This may mean that the analysis programs should
be left as floats (since the analysis files are floats, this is
pratically a requirement).  But I just don't see a major downside to
doubles, except for the time it will take to do the conversion of the
source code.
-- 
Mike Berry
mikeb@nmol.com