What is not reflected however is the sheer number of votes that people tend to cast in the US - often voting on dozens of different measures.
The UK process does not have this aspect - at worst there may be a Westminster, country-specific parliament and city or town council on the same election day, and they are handled via separate ballot papers.
It’s literally an ‘embarrassingly parallel’ problem.
A hundred people can count ten times the ballots that ten people can, and then merge their results.
The first is that there are a lot of reduction steps where we're doing many to one communication. The more nodes you have the more communication there has to be and this is an issue.
The second issue is that if we (naively) assume that there's a consistent error at each step, that the error will grow in each gather step. So if each worker has a 0.5% error rate, each gather has a 0.5% error rate, then our error rate depends on the number of workers and gather steps.
For what it is worth there is at least a gather step at each poll station, then each precinct, then the state, and then the country. It is also likely that these all get worse as the election is closer and there's more pressure on.
US elections are also operated entirely regionally; the largest US election is that in California, which is smaller than the UK.