I made an estimate and while it was higher than I expected, the increase in power is not anywhere near what you mentioned.

I'm using the value of Cd * A from here (0.6 m^3): https://www.sheldonbrown.com/rinard/aero/formulas.html

That page also indicates that the vast majority of the energy expenditures on a bike are due to drag.

The fraction of energy needed to get up to speed divided by the energy expended due to drag works out to: (mbike + mhuman) / (Lblock * rho * Cd * A) where mbike is the mass of the bike, mhuman is the mass of the human, Lblock is the length between stops, rho is the mass density of the air, Cd is the drag coefficient, and A is the projected area. I can post the derivation if anyone is interested.

Assuming an 8 kg bike, 90 kg human, 200 m between stops, and an air density of 1.2 kg/m^3, stopping adds 68% more energy.

This is dependent on the weight. If I crunch the numbers for my own weight, stopping adds about 56% more energy.

So, this is higher than I expected, but I don't think it's so much higher that it justifies running red lights.