That looks like too large of a difference to not have an explanation to me, like you're not actually comparing the same points.
I know absolutely nothing about the format but I notice inside the function the data are encoded as integer w/ apparently some scaling; perhaps there's something going on with the latter as integer values should be identical.
The values are taken from the file as
tmpdata(recnum).data{ii} = fread(fid,hdr.samples(ii),'int16') * scalefac(ii) + dc(ii);
where the scalefactor and DC offset are computed from some header info.
I think I'd look at the input in the file more carefully before just proceeding here...
