But first, you changed R1 to a linear resistance. :) After thinning, it dissipates much more power than it had before.
As power across R1 increases, it's resistance increases nonlinearly. As power across R2 decreases, it's resistance decreases nonlinearly.
I'm not going to do the arithmetic, but it's a PTC series circuit with a constant voltage across it. Is it plausible that net power dissipated by this network remains fairly constant despite the fact that the ratio of power dissipated by R1 is inversely proportional to the power dissipated by R2 over a range of say 3% of P2? I think so.
(...)
I ain't so sure. My cheapo SPICE simulator does not have a PTC thermistor component, else I would ask the computer for an answer.
--Winston