We obviously have a huge variation in spread. Indeed, the choir is evidence of superspreaders who produce many cases, and we have an average R0 of 2-3. Singing loudly in enclosed spaces with many people is an example of a contact network structure that causes greatly increased susceptibility and contagion. Indeed, it proves the whole point!

But, the belief that herd immunity = 1 - (1 / R) is based on the assumption that those infected have equal susceptibility and equal probability of spread. This is a fair assumption when it comes to a vaccination program which will administer doses not correlated to susceptibility and contact network structure. It is not a fair assumption when it comes to actual spread of disease in the wild, which will preferentially spread in the more susceptible networks.

Reality is certainly better than this; the question is how much better. There's been some quality analysis of time-series data that implies the threshold may be 20-30%.

I believe that New York City's quick decline in death rates compared to jurisdictions with similar or stricter controls but lower seropositivity may imply that ~20% exposure is enough to significantly attenuate spread.