First, you should check both creation dates of the duplicated elements and blame the author who created the duplication without clean-up. If the duplicated data is coming from the external data import, then you should check other concerned areas and consider reverting the import if itself and possibly revert it when it is not done in the state of the art (see our wiki about [imports][1]). [imports][1]) (perhaps the import was decided in this way and conflicts are solved manually by local contributors, check this point with your local community first).
If you consider that the amount of duplicates is manageable and the imported data is valuable, then you have to decide case by case which of the duplicated way you want to keep and which one will be deleted.
deleted:
In the [example you are pointing][2] for instance, the imported waterway is qualified as a 'stream' and 'name' is missing where the manually surveyed version says it is a 'river', 'width' is '20' meters and the 'name' is present. Based on this example, I would say that the imported data quality is very poor (name missing), badly managed (they did not check for duplicates) and use wrong tags (20 meters wide waterway is a 'river', not a 'stream'). .
In your position, I would check other examples of this imported data and revert it globally if there is no real added value but this should be done only after some discussion and consensus inside your local OSM community.
[1]: http://wiki.openstreetmap.org/wiki/Import
[2]: http://www.openstreetmap.org/?lat=49.6288&lon=15.49187&zoom=17
