>Probably most major commercial ones before 2000 were.

I find that hard to believe. For instance, what makes you think they got speculative execution more correct before 2000 than after? At least it seems impossible to get rid of the timing side-channel since the whole point of the exercise is to change the amount of time it takes to run the program.

Surely you'd have to go back to at least 1980. I suppose one could start a production line of 1970s supercomputers for personal use, and put it on the programmers to figure out how to parallelise everything, but it would be very painful and expensive.