In reply to F6ENO:
I assume you are using the summits.kml file which I produce.
The coordinates in this file are derived from the SOTA database, which in turn are supposed to correspond to the respective ARMs. I do any necessary conversions as accurately as I can, and round the final result to 4 decimal places of degrees. However the accuracy is in general nowhere near as good as that precision might lead you to expect.
First of all, ARM coordinates are normally for identification purposes only, and there is no claim that they represent precisely surveyed positions. The inherent precision of the raw data may not be very good; in the UK for example the coordinates are based on 6 figure grid references, so a coordinate represents a 100m square.
Even though all of my processing is automated, it is clear that there has been a lot of manual processing of this data, and it would be surprising if some errors had not crept in. A few of the more blatant ones have been corrected.
On top of all of this, there is often a very real difficulty in knowing where the summit actually is. A map will not tell you accurately; contours are not normally accurately surveyed but are usually drawn from aerial photography data. Different countries are likely to survey to different standards. Accurate determination of the highest point on a flattish top can be extraordinarily difficult.
You cannot rely on Google Earth to tell you where the summit is. The 3D model that you see is made by combining the satellite and aerial photography data with an entirely separate data set of height data based on radar measurements. The height data is recorded on a much coarser grid than the images. What you see is the result of wrapping those images onto a 3D surface which is largely the result of interpolation. It works very well to give you an idea of the terrain, but there is no way that it can be used to determine the highest point.
Taking all of these factors into account, I think we have to regard the Google Earth file as just a convenient indication of the approximate location of a summit. It would be good to have more accurate data, but to do it consistently would be a lot of work.