There’s an idea that is almost as oversimplified, makes a big assumption, and comes up with a much lower percentage: suppose that each person is just as likely to interact with any other given person, but that people come in two types: susceptible and naturally immune. Then the initial R0 in fact represents a more contagious disease than the first model (as the probability of infecting a susceptible contact needs to be higher for a given R0), but the disease stops increasing exponentially at a lower infection rate. In particular, once the entire susceptible population has gotten the disease, it’s over, and that’s less than 100% of people.
Both of these models are hugely oversimplified, and neither one is likely to be fully correct.
Oh, I see where the 68% comes from - https://www.nytimes.com/2020/07/09/nyregion/nyc-coronavirus-...
That's from one clinic, for people seeking an antibody test.
Anyway, the first link I mention is the broader dataset from all the clinics like this, so it helps smooth out anomalies. But, the big caveat in all of this is that these suffer from selection bias - it's not a randomized sampling of people in each borough. It's still useful data, but it's far from being able to tell one the incidence rate to a reasonable degree of precision. Still, many reasonable inferences one can make.
Even in your first link, the highest percent positive is the Bronx reporting 33%. Not sure where you're getting 51% from, but it very much seems like you're trying to cherry pick data to support your argument.
If NYC is that high - I missed that - then maybe parts of NYC are actually close to the target.