You've entirely missed the point I was trying to raise, Philippe. It was not the ability to do private normalization of private use characters that I was calling for, but making it easy to do. Private Variation Selectors and Private Combining Marks with a non-zero combining class that have appropriate default properties are not necessary to enable normalization according to a private use version of Unicode. However, they do make such private normalizations easier to achieve, as they are not dependent upon the private use agreement changing the defaults of properties that affect normalization.
Besides, I don't think we want to encourage having font-based overrides of the properties that affect normalization. That could would raise all sorts of potential security holes unless such changes were both restricted to only the PUA and PUA characters were not allowed in secure applications. (The latter may well be good idea unless the private use that is desired specifically calls for such characters.)
