1) Punishments get scaled by the judged likelihood of guilt, i.e. judge says there's a 65% chance Bill is the killer, Bill gets 65% of the punishment.

I think this would be very hard to make work as long as there remained a significant component of human judgment in determining probability of guilt. It seems much more likely that instead of working as intended, most of the time the people responsible would rationalize whatever probability of guilt would result in the level of punishment they wanted. You'd get very distorted probabilities out of this. Certainly we see that people do this from the historical cases mentioned in the article, when death sentences were required for many crimes, so juries adjusted accordingly.