In particular, this setup was actually I/O bound and there wasn't more I/O to be had.

No more I/O to be had from that particular NFS server. If your mounts are distributed over 4+ servers, then you have potentially 4x the available operations available.