Please send me one test file!
I guess it is related to multiple 'comments' being present in your file.
Sorry, but now it get's into ID3v2 details (and bear with me but many MP3 tagger treat things differently here)...please see here:
http://id3.org/id3v2.3.0#Comments
There may be more than one comment frame in each tag, but only one with the same language and content descriptor.
ProppFrexx by default uses the language as set in your OS, which is the current culture being defined in your Windows settings (I guess this will be 'ell' for 'Greek' in your case).
And it uses the first found 'content descriptor' present for that language (resp. an empty content descriptor when writing).
When reading a comment, ProppFrexx is a bit more 'relaxed' and might even return another (the next best found language) if the language of your current culture is not present.
As such, it might happen, that a file contains a comment in for example in Italian or English, but not in Greek - so ProppFrexx uses that comment (as a fallback).
However, when ProppFrexx writes to the file, it uses YOUR language - as such a 2nd comment might be created.
This is totally correct - as the next time ProppFrexx reads the file/TAG - it wouldn't use the 'fallback' - but the correct 'language' and as such also returns the correct comment.
Unfortunately (when you for example look at Mp3Tag) - they hide this details from you.
It might get even worse: As per language you might theoretically also have many comments for different content descriptors.
ProppFrexx might read the 'first' found, but write an empty content descriptor.
Luckily most 'taggers' I know all use an empty content descriptor - so this might unlikely happen.
But many 'taggers' I have seen also just ignore the 'language' code and write comments with an empty language code field!
This is simply incorrect and not according to the ID3v2 specs (see above) - but unfortunately needs to be handled in the best possible way.
Which is 'reading' comments in a 'relaxed' way - but writing to it in the 'correct' way.
(I could have decided to 'delete' those previously incorrect 'comments - but I have decided to leave them as they are, since another application, the one who wrote the comment incorrectly might still relay on its presents)...
In your case I exactly assume this...
Yes, complex - but I believe correct!