Yes, increasing width can help with noisy data. Must remember that width delays the detection to approx width/2 points past the actual peak if you're using the pt by pt vi (or peak detector.vi in realtime processing).
You could also set the valley threshold to something like (Mean-300) if your data is consistant from run to run.
Or you could get into more elaborate data smoothing before you pass the data to the peak detector. If you haven't already checked it, the following link may be useful:
http://zone.ni.com/devzone/conceptd.nsf/webmain/55181A206A523FD186256ABE00626EA9?opendocument&node=dz00000_us#7
=====================================================
Fading out. " ... J. Arthur Rank on gong."