It's not just that, the one image of 88C with 80% fan speed or something was still only 2300rpm or something, a 290 at the same temp and same fan speed would be at 4500-4600rpm or something in that range. THe 290 fan is quiet below 2500rpm, a bit too loud 2500-3500 and horrible from that to max of 5700rpm.
Temps don't matter all that much, you have to factor in the noise being created to produce a given temperature. Same temps as a 290x at the same noise level is one thing, if it's half the noise or less then it's a completely different comparison.
As you've also pointed out and I keep trying to point out, a new node doesn't do anything directly for heat reduction. You're doubling transistor density while halving transistor power(average aim for a new node anyway), that means pretty much same power output per mm^2.
Power output per mm^2 is the thing that will have the most effect on how easy it is to cool a card. After that you just need a heatsink large enough to dissipate the heat with a given airflow, which is relatively easy to achieve.
The 290x was precisely so hot because it has one of the highest W/mm^2 ratings of any core in years. The 1080 is 40% worse(higher) than a 980ti. The RX480 is around 980ti levels(about 0.41-43W/mm^2, 290x was 0.63W/mm^2 or something and the 1080 is 0.57W/mm^2.
stress testing tends to produce much higher temps and stress, but again even then the numbers I've seen suggest dramatically lower fan speeds than compared to the 290 at similar loads, half the fan speed which means likely 1/3rd of the noise or less.