For any distribution P(x), defined by a standard deviation s and a mean x0, Mahalanobis distance of any point x in this distribution is |x-x0|/s. Obviously, there will be some values of x for which this will be < 1, and others for which it will be >1.
My interpretation is that for values you are mentioning, sparsity is exponentially favored. Some images showing lp spaces in the Wikipedia are quite illustrative so far. Unit "balls" are conceived for p >= 1, thus "astroids" do so for 0 < p < 1.