That conflicts with OP's description saying their Steam downloads were 20% slower, which presumably was tested before hitting any soft transfer cap. I'd guess

That conflicts with OP's description saying their Steam downloads were 20% slower, which presumably was tested before hitting any soft transfer cap. I'd guess the goal of traffic shaping on CPE would be to keep upload bandwidth on the cable segments less congested. Or perhaps conntrack is easier to do in a distributed manner?