It feels like a hawks and doves style ESS (evolutionary stable strategy) problem. It appears to me that the strategy that eliminates the overall negative affect for the group (avoid creating phantom stop signs, by gapping or "always in the middle" strategies) is necessarily unstable.
The benefit to any single driver who does not follow that rule becomes increasingly high as more and more others do follow the rule. Even to the point of ignoring a beeping car I should think. I certainly feel the negative effect driving behind someone who creates a gap as more and more cars merge in front of them. In the pathological case the driver who insists on a safe interval between cars will make no progress at all.
1: (CGP Grey) https://youtu.be/iHzzSao6ypE
Of course you must maintain speed. The point of a gap is to allow for adjustments without the hard breaking that would create a phantom stop sign. That is the overall intent of the system. Move the most people as expediently as is safe from point a to point b. The best way to do this is to avoid speed differentials, being slower then ambient traffic speed creates a speed differential.
This is more of a thought experiment to show that you never have to keep slowing down to zero even if for some reason people continually enter your gap. I barely observe people entering my gap at all in practice, because all lanes are generally going the same speed so there's no benefit to it.
That lowers the throughput (number of cars that complete a trip in your lane) by ~ 10%. Also, the faster lane is now probably moving closer to your speed behind you than the 50mph you observe (because you are causing cars going 45mph to cut off cars going 50mph that lane).
The arrival rate of cars won’t change because you chose to go slow. So, if the road can handle 10,000 cars per unit time with normal congestion, it can now handle ~ 9000. That means per unit time, there are 1000 new cars sitting in the stop and go backlog that you single handedly created by creating the gap.
This has been shown time and time by traffic simulations. It only takes a small handful of slow drivers to create massive backups.
Also, the number of traffic fatalities is proportional to the difference in speed between the fastest and slowest car, so, statistically speaking, by being the slowest car, you are murdering people.
With the pathological case being so benign, the real life experience is even better. You might get a car moving in front of you once a minute, and usually a few seconds of slower accelerating or coasting is all you need.
To me the thing that really slows down traffic is people not moving left to let onramps merge. I think that's because they are afraid of missing their exit because no one is leaving gaps.
I just drove a Ford Edge that had this. It is a little overeager about what it thinks is an impending collision, but it works alright.
The adaptive cruise control is a mess though. In theory, it is supposed to match your speed to the vehicle ahead if they are slower, so you don't have to pay attention. In practoce, it has the potential to be a great contributor to congestion. With dumb cruise control, you can push slowpokes to get out of the way or at least speed up; the adaptive cruise control seems to match speed at a distance where the driver ahead doesn't get the signal that they are obstructing things, and they continue bumbling along.
As for parkways, the better flow is because they don't allow trucks. Variance in the speed of cars is what causes traffic. Obviously this is true at on/off ramps. There is no way X95 can take two lanes of high-speed flow of cars into their already packed flow. For parkways with a lot more hills, you can still get slowdowns but these happen almost exclusively just before big hills and at exits to bigger highways (and occasionally at busy on ramps with short merge areas).