Wow... That came out much better than I would have expected! There is no way I could have done that on my printer. With that said: It doesn't push things any harder to go past 85 the way the code is written.

z=83; difference() {
sphere(r=25, $fn=100);
translate([-20,-25,-z]) cube([60,60,60]);
}

It is taking a sphere with a bottom point of z=-25 and pushing a cube with a top point of z=60 down, and subtracting it. Once you push the cube down 60+25, you are no longer subtracting the bottom of the sphere off of it. If you use the code snip in this post with the sphere(r=25,$fn=100); it will tell Open_SCAD to have a much greater resolution on the surface of the sphere. And the amount of adhesion you get when you 'push things' will (in theory) go to a single point. Probably you will have problems as you approach z=85.

Those first layers did curl up, right? They look a lot thicker than the rest of the layers.

If you want to experiment.... You can use this snip of code with one extra line in it. It will attach a 1mm panel under the sphere so you don't need support. The plate should hold adhesion to the plate and as you approach z=85 you will be bridging out at almost a 90 degree angle.

z=84.75; difference() { sphere(r=25, $fn=100);
translate([-20,-25,-z]) cube([60,60,60]);
}
translate([-25/2,-25/2,-(z-60+1)]) cube([25,25,1.1]);