Hence why you set your QoS profile to less than your typical connection throughput. That way you control the QoS, not your upstream. It's somewhat inefficient, but it works when you're saturating your connection and want to make a VoIP call or something else more sensitive than bulk traffic.
I agree you are both technically correct in your own sense. In my experience downstream traffic shaping with m0n0wall and pFsense I really had to cut my "max" down to 75% or less than my actual downstream was before I started to notice anything.
Since cable modems generally have powerboost and latency increase is minimal when fully loaded downstream in the end it makes little sense to have downstream shaping.
Perhaps DSL is different (or cable now with hard caps) but I downstream shaping is definitely iffy...