I don't remember seeing that predicated as a low-level I/O savings — it was always things about being easy to cluster or avoiding the performance overheads of ACID. The pitch was usually “it's easier and you don't need a skilled DBA”.
What I heard was "You can practically scale indefinitely". Which I guess is more true of NoSQL. As is I/O savings. But it's not something I've ever heard it advertised for.
Performance was an imposed callsign of Mongo, the reason everybody started using it was the declarative schema stuff and so the dev didn't have to know how to start a terminal, which was a not-insignificant hurdle. It doesn't have significant performance differences over other simple KVs. "Faster" was primarily a measurement of learning curve.