What is happening here is that you are changing
the sampling rate in your Csound orc, while this
information is not getting to the DAC.  It remains
playing at some fixed rate.  So if you make more
samples per theoretical second of a piece, the DAC
takes longer to spit them out, and thus lengthens
the piece, and lowers all of the pitches.

If you are using a Linux system, I recommend 'vplay'
as the command line sound file player.  If you choose
'-W' on the Csound command line to be sure that you
are generating a 'wave' file, then vplay will pick 
up the sampling rate from the header of the sound 
file, and pass that information on to the Dac, so that
the pitches and durations of the sound remain constant.

Then you will want to choose the highest supported
sampling rate where sound quality is the most important,
and disk space and compile time are the least important.
As the latter two conciderations increase in priority,
the sampling rate must be lowered, sacrificing some of
the quality of the sound, most notibly high-end 
frequency response.

Toby

	-There otta be a law-
> But this seems to depend (even in the latest version (3.781)) on the
> sampling rate.  For example, when I do the little example score in the
> manual, but set a sampling rate of 44.1kHz, the whole `piece' is much
> longer than 2.5 seconds (seems to me it's about 4.41 times as long...)