Correct, but without any real traffic happening, apart from idle background chatter.
Correct, albeit in both cases - 4+1 ports in case of the nbg6817/ ipq8065, 3+1 ports in case of the gateprotect, 1 port in case of the ASRock, everything the devices have to offer (link up, idle background chatter, but not trying to push the device).
So is baytrail-d, launched Q4'13 based on a 22 nm process, the comparision isn't that far off (ipq8064 was certified by the FCC in mid 2014, as a revised spin-off from the APQ8064 smartphone SOC, which was released in mid 2012).
Not if they have 5GBASE-T or 10GBASE-T ports, far from it - the simple ipq8071a based xiaomi ax3600, yes (~6.5 watts, but that bails out at routing ~600 MBit/s, without SQM - while the j1900 can easily route 1 GBit/s at wirespeed without waking up and with sqm/cake up to ~830 MBit/s).
EDIT: the Q2'08 vintage Atom N270 (1* 1.60 GHz + HT) on a 45 nm process also manages to route ~600 MBit/s (without sqm), at ~28-29 watts idle (Intel D945GSEJT, 12V/ 5A Notebook PSU, 1* r8168 NIC, 2 GB DDR2, 500 GB 2.5" spinning rust).
Maybe, but that would only be relevant at full load - not for my idle- or long term measurements, the ARM cores (2* KRAIT300/ cortex A15 @1.7 GHz in case of ipq8065, 4* cortex a53 in case of ipq807x (ipq8071a: 4* factory overclocked to 1.4 GHz, ipq8072a+: 4* 2.2 GHz; 4* 2.0 GHz (2.42 GHz burst) in case of the j1900) clock down to the minimum for those tests anyways (and remain there for most of the day).