You are right that your numbers are over-estimates. They would be "correct" for a plate with no resistance to bending, which was only supported along two edges and heated across its full width between the unsupported edges.
This is a hard problem to produce a simple "formula", because in the real situation the bending will be reduced by the stiffness of the glass, and because the deformed shape is not part of a circle but a bulge in the center of a square. In real life there will be some transition between the temperature of the heated part and the surrounding "frame" as well, and not a sudden temperature change.
The best way to get a theoretical answer would be to model this with finite element analysis. Alternatively, learn from other people's practical experience, which is quicker and cheaper than learning from your own practical experience!
FWIW the simpler case of a circular plate was apparently studied as part of a PhD back in 1960, with this report as a spin-off: http://naca.central.cranfield.ac.uk/reports/arc/rm/3245.pdf