Well, XFCE resource usage grew with GTK3, probably :(. But there's something off in their test methodology. They claim to measure idle RAM usage, and there's 2x difference between tests, on same distros..? (for comparison, my Arch XFCE install takes 480MiB after boot)