Most power generation facilities are natural gas fired, using large aero derivitive gas turbine engines (essentially the same engine that is in the 747 - LM6000 vs CF6 for example) with a combined cycle steam turbine to capture the energy from excess heat. This arrangement has a thermodynamic efficiency of 60%. Even with electrical transmission losses, the efficiency is still far better (1.7X) than having the power plant located under the hood of the car.
Not to mention, policing one large facility for compliance with emissions is much easier than trying to monitor every single one of millions of cars on the road.
Even with a car, you aren’t necessarily regulating millions of individuals - primarily just manufacturers. I suppose there is the odd case where “old joe” removed his catalytic converter and is polluting more than others but that is probably rare.
I don’t get your point on centralization however - more efficient but less robust (just like in software).
For necessarily centralized industries like car manufacturing the regulations are written to protect the wealthy incumbents competition more than to protect consumers.
It's a different metric but it is generally accepted that F1 cars have reached an overall thermal efficiency of 50%, which is cool. This is taking into account energy recovery from kinetic (regenerative braking) and thermal sources (from the turbo).
note that this is true for spark ignition engines, but not all ICE, diesel engines can reach higher efficiency and there are even real diesel engines with almost 50% efficiency[1], obviously not in cars though.
I fell down a rabbit hole and found this link, which gives 46% for the theoretical limit for the efficiency of the internal combustion engine.
https://physics.stackexchange.com/questions/98966/maximum-th...