While this is technically true (the devices themselves are cheap and effective), the data required for them to work well doesn't exist. If the data exists at all, it is usually horribly out of date. So in practice there will be many cases where these devices limit you to the 30mph from last year's big construction project. Then you get rear-ended because nobody expects you to slow down 50mph for no reason at all.
Which is to say, if you were to build out a system that limits speed based on some authoritative database of speed limits, then suddenly there's an incentive to make sure that database is actually correct (where there was no such incentive before).
Maybe let's start with residential streets only, using the residential speed limit. And let's say only those residential streets that are at least 100m away from a faster street.
I'm sure it will be possible to find a solution that eliminates false positives.
And in the worst case - it's just about acceleration. It's not like the car would abruptly break.
Even in this rare fictional scenario, I don't agree that the costs outweigh the benefits. There are many vehicles that cannot go above 30 kmh and somehow they don't get rear-ended either. Plus, mid-term, the other cars will also have the speed limiter installed.