I think that might be premature optimization to worry about finding random points in an n-dimensional hypersphere where n is a huge number.
A good intuition for why this happens is that the distance from the center of an n-dimensional hypercube to any of its corners is `r * sqrt(n)`, but the distance from the center of a ball to its surface is always just `r`. So if `r` is fixed, and `n` keeps increasing, the corners get "spikier" and farther from the center. Very quickly, almost all of the hypervolume of a hypercube is located in these remote corners, and almost none of it is in the ball at the center.
I interpret the comment as saying "it's premature to worry about (finding points in n dimensions where n is large)". This interpretation implies n is not large.
Not "it's (premature to worry about finding points in n dimensions) where n is large". This interpretation implies n is large, but that you don't care.
It's not just a geometric object that has limited practical value for d>3. It is the set of points where the norm is unity, which is a pretty fundamental concept in all sorts of places.
In terms of actually solving to proble you just want a vector random distance and a function that to map random distance to actual distance. But if you never deal in 6+ dimensions that's a waste of time.
Which is where that method was used.
When your problem is n=3, anything larger than 3 is “huge”.