Coordinate operations may be slow, especially when datum shifts are involved. When not using datum shift grid files, the most common method is geocentric translation. The Molodensky method is a little bit faster at the expanse of accuracy, but that accuracy lost may be acceptable for example if the scale on a map is not sufficient for seeing the difference. At very small scale where each pixel is 10 km wide, we can even skip the datum shift operation completely. The question is where to put the threshold. 5 meters for Molodensky, 10 meters for Abridged Molodensky, 3 km for none? This need to be measured, for example by trying all geocentric translation operations in the EPSG database. Once thresholds have been determined, the DatumShiftMethod.forAccuracy(double) method should be updated accordingly.