YEAR DEATHS POPULATION
1999 5,508 24,099,614
2000 5,561 24,272,121
2001 5,513 24,515,654
2002 5,943 24,742,879
2003 5,582 25,019,050
2004 5,564 25,510,811
2005 5,234 25,839,430
2006 5,186 26,116,159
2007 4,986 26,327,320
2008 4,067 26,408,974
2009 3,514 26,378,690
2010 3,159 26,206,325
From 2000 to 2009, the # of traffic-related deaths per 100,000 resident population dropped by 26%. Assuming the decay is exponential, that's about 3% per year. In contrast, the # of teen deaths dropped by about 4.5% per year.
So we can reasonably conclude that teenage auto-related deaths are falling at a faster rate than overall auto-related deaths.
Is the fatality rate for all inexperienced drivers dropping the way it is for teenagers? If not, that suggests that the apparent improvements aren't so significant as they seem. In particular, social and regulatory changes that make it more difficult for teenagers to drive, but not for young adults may only shift what age group the problem affects.
If you reduce the driving privileges of people with blue eyes, I'm sure that the auto-related fatalities of blue-eyed people would decrease.
I'm not sure of the correct way to measure the success of driving restrictions, but I would suggest that it is something other than this type of statistic.
If there exists a mushroom that tends to be lethal in even small quantities and its usage in recipes is restricted, then the fatalities caused by it will go down.
You might disagree with the method on deontological grounds, but you can't really deny that it is functioning well on consequentialist grounds.