This is eye-opening, and the effect on KNN, for example, is easy to predict, but should the limitations of Euclidean distance in high dimension be a reason for concern in the very common application of multiple regression?