Those highlights match my field experience.

I've found that ensembles aren't necessarily more accurate for individual forecasts per se, but in terms of aggregate errors over the long run, they end up being less wrong. (bias-variance tradeoff)