No reason why 85th percentile speed limits should reduce accidents

The "85th percentile" is a buzz phrase that has been around for a while now in speed limit-setting circles. The 85th percentile speed can be defined as "the speed at or below which 85 per cent of all vehicles are observed to travel under free-flowing conditions past a monitored point". In other words, it is the speed limit that most people observe on a given road, regardless of the posted speed limit.
There is a movement in traffic management circles to use this as an official speed limit, on the basis that few people (15% presumably) are likely to exceed it, that it has de facto been proven that people can drive safely at or below tbat speed on any particular road, and that such a speed will minimize the number of crashes and fatalities.
Now, this makes absolutely no sense to me. A speed limit should be designed to constrain the speed of vehicles to a speed considered safe for that particular road, taking into account pedestrians, curves, intersections, etc. It should not be based on the speed that people happen to want to drive. And my initial thought was that if you increase a road's speed limit to the 85th percentile speed, then people will just drive a bit faster, and the 85th percentile will then be higher still. And we know that increasing speed limits increases accidents, particularly fatal accidents - whatever you might like to think, and however many anecdotes of Germany's autobahns you might like to quote, this is nevertheless a fact, even in Europe. It seems like common sense that increasing the speed limit on a road that was designed for a lower speed limit will lead to more accidents.
And yes, it does seem that the experience in the US bears this out, as does evidence in Canada. Sometimes common sense is just the way to go.