Why not random matrix draw from the 1.0 / n_feature scale?

As can be seen from here: https://github.com/scikit-learn/scikit-learn/blob/fd237278e895b42abe8d8d09105cbb82dc2cbba7/sklearn/random_projection.py#L163

It says: The components of the random matrix are drawn from N(0, 1.0 / n_components).

I wonder why the scale is not according to the Dimensionality of the original source space. But it is related to the Dimensionality of the target projection space, which means the higher dimension the target space is, the smaller the l2 of random matrix vector should be.

It doesn't really make sense to me. I think it should relate to the Dimensionality of the original source space and draw from the 1.0 / n_feature.