It seems to me that the guy's "vanilla" approach is opening up a new authenticated connection for every email, whereas his proposed "new" method is reusing the

It seems to me that the guy's "vanilla" approach is opening up a new authenticated connection for every email, whereas his proposed "new" method is reusing the same connection for multiple emails. Not surprised it performs faster... Presumably the workers are spending 99.9% of the time blocking on network IO, so it's not a great example for true parallelism; a single thread managing many connections asynchronously would probably be the fastest!