Ah. What a pisser. I wonder if you could make Text::CSV_XS work by reading from STDIN... If so, you would just filter out all the BOM characters before feeding the data to your script:
perl -CS -pe 'tr/\x{feff}//d' < source_file.dat | your_csv_parser ...
Either that, or else redirect the output of that one-liner to create a cleansed version of the DAT file that has all the BOMs stripped out, and use that "bastardized" version of the data as input to the parser. (I assume that getting the data parsed is more important that preserving its obtuse fixation with BOM characters.)
