> They'll be scaled such that the point not on a plane will evaluate to 1 when plugged into the equation.
Evaluating a plane equation with a point will give a signed distance from the point to the plane. However, unless the equation is normalized, the distance could be scaled in unusual ways that are less than useful. I think a better way to say this is that a point who's shortest straight line distance to the line is 1 will evaluate to 1, after taking the absolute value.
The LRBni ISA had a whole set of instructions designed to take advantage of signed distance fields defined in both plane equation & barycentric forms. (That's what the bit mask for the lanes was for, in part.)