Thanks a lot for the various replies.
There are several things to unpack and the discussion is quite interesting.
Regarding the risk of "memory overload", the
wiki states that "(technical) increase of volume will increase requirements in processing power,
but Moore's law and cheaper hdds every year are always there. This might be more complicated when we will speak about geospatial queries rather than simple linear read/write patterns." and there isn't any clear direction against micromapping. We are clearly dealing here with an arbitrary decision by one user. Similar to the principle of not "mapping for the renderer", I don't think that anyone should "map for the server", especially when acting upon a personal intuition in contradiction with the wiki and particularly before damaging other people's work.
Regarding the possible imprecision, as Danilo pointed out, the appropriate way to deal with it would be to correct it / shift it, not to damage the data.
Finally, I understand that different people have different more or less valid prejudices (including of course me) regarding openstreetmap. The healthy attitude consists in mapping differently, not attacking existing precise data. The argument thant "Valais still has so much potential" very much also applies to the data attackers. I would like to point out that this argument is somewhat childish considering the amount of "useful" data that I have contributed in Valais as well as in the developing world.