All right, how about this:
The small beam on the right is subject to pure bending, with a square cross section and the larger beam on the left is now just a wall that the small beam is attached to. So for a predicted failure model:
σmax = 6M / (h^3)
where we could set σmax = yield stress of the material
M = Fmax * L
L = beam length
h = height of square cross section
Rearranging, I get:
Fmax = (σmax * h^3) / (6*L)
What do you think of this?