It would be interesting to see the actual discrepancies that your users report - are they completely random or is there a pattern; are the discrepancies small or large; do the discrepancies start at a particular size of file; is there a difference between text and non-text files (I could go on ...)

cheers