According to this:

https://dqydj.com/weight-percentile-calculator-men-women/

That female dummy would have been in the 24th weight percentile, while the male in the 57th female weight percentile.

Which is a good argument for using a range of test values, not for using less representative values.

Cars aren't cheap

Compared to the cost of designing a car then manufacturing hundreds of thousands or millions of them, a handful of cars are pretty cheap.

Depends on how far down you want to go with all the people sizes in all the various combinations of positions. I imagine the cost-benefit is based on the marginal change of information you get by changing the dummy size/shape, and based on their revealed behavior, they don't think more sizes add more information.

When I worked at Toyota, they used "number of Highlanders" equivalence when shaming people for not picking up their large format paper schematic prints.

exacty, they shoud be testing the heaviest, tallest, and the shortest, lightest.

But how about a tall and light person and short and heavy person? The permutations are endless. Not arguing against testing, I think it's great to get more test dummies in the seat, but I don't know where/how one should draw the line.

If you're testing those two dimensions, there are four corner cases. It may be that for some reason somewhere in the middle is a worse case, but the extremes should certainly be tested.

I think somewhere along the lines of 20, 50, and 80 percentile would work.

Across how many dimensions?