There are deeper mathematical principles of measure concentration that make random projections work. One famous result is the Johnson-Lindenstrauss lemma stating that a set of n-dimensional points can be nearly-isometrically embedded into a log(n)-dimensional space by means of a random projection.