I'd argue that #2 (more efficient processing) is not the case, and is highly dependent on implementation.

@silverpill@mitra.social @raphael@mastodon.communick.com