Why must the classification of signal or noise be binary? Redundant addressing might legitimately fall between the two. Even if you do consider it noise, it is the most easily ignored noise possible, always in the same place with the same format.
I believe the downsides of this feature outweigh the upsides.
Edit: I believe the proper categorization of the @ tag is not pure signal or noise, but redundant signal. A redundant signal may be removed without changing the meaning of the message, by definition. However you will find that human speech is full of redundancy, because it enhances communication and improves understanding. I would argue for its retention in this case.