Try using iconv instead of dos2unix to convert the Unicode character encoding scheme of the files. The gremlins in the text might be something like improperly unpaired UTF-16 surrogate characters. I would expect iconv to handle these anomalies better than dos2unix. (It should at least warn you about them in the default case.)

Just a thought…

UPDATE: Another good tool for diagnosing peculiar and elusive character encoding problems is BabelPad.