Motivated by the conversation here, I produced this graph showing the per-stall efficiency gains compared to a 2-stall site, for different probabilities of blocking (1% to 10%).
What this chart means, for example, is that a 20-stall site with 10% chance of blocking is 3 times more efficient on a per-stall basis than a 2-stall site with the same 10% chance of blocking. So the 20 stall site is 10x bigger than the 2-stall site, with 3x better per-stall efficiency, meaning overall it could cope with 30x the rate of car arrivals as a 2-stall site and still achieve 10% blocking.
It's a bit counter-intuitive that dimensioning to a lower level of blocking means the rate of efficiency improvement with more stalls grows even faster than at higher blocking levels.
But that doesn't mean DCFC providers should design for 1% blocking because the rate of car arrivals will dictate how many stalls they need to achieve that 1% blocking. The lower the blocking percentage target, the number of stalls required to serve a given rate of car arrivals will blow out exponentially (well, Erlang-ly
). So the amount of money needed to achieve lower and lower blocking percentages escalates rapidly.
You can see from this chart that as sites become very large (> 30 stalls) the growth in efficiency per stall starts to become fairly linear and asymptotes. So lots of bang per buck to expand sites beyond 2 stalls if the rate of arrivals demands it, but not to go from 30 to 60 stalls.
This chart is based on the Erlang-B formula, which assumes any request not immediately served (blocked) is lost forever (i.e. driver just leaves) and there is no queuing. Erlang-C formula is used to model arrivals with queuing.