In cinema, photography and game design, it's fairly common-held knowledge that certain type of perspective and movement can cause objects to seem out of scale with each other, or simply to receive less attention from the viewer.
The solution to such a problem is to play with scale empirically until things "look right". I've not heard of an exact formula or precise set of numbers for it.
So to answer your questions, playing with scale to make set compositions look better or to alter the audience's visual attention is a well known trick. The 33% stipulation is just something this guy picked up which works for a particular scenario.
I agree with Patrick Hughes though, if you're not sure if it will work with you, try it yourself. Just don't be afraid to try values other than 33%