#4: With all the air (more or less) removed from the dough, the ambient air pressure is free to compress it. Imagine the wall of an empty container: the air inside is at ambient pressure (let's say 1 atm, or about 14.7 pounds per square inch) and the air outside is also at 14.7 psi. Now remove the air from the inside. The pressure of the ambient air still acts on the wall: 14.7 pounds on each square inch. Hence, pressure; hence, compression.

It might be more intuitive to imagine something like a sponge in a bag; as the pressure differential increases, the net force on the bag's surface "compresses" the sponge like a spring.

The Wikipedia article on "Vacuum" is pretty good.

I think commenters are missing the distinction here, which Egypt Urnash pointed out all the way up in #1. Our brains and eyes--correctly!--perceive that A and B are different shades on the modelled object. One is a dark square, one is a light square. (Compare adjacent tiles to confirm this.) We don't care what color paint is needed to paint the "picture", we care what color paint is needed to paint the chessboard or cube. So the light square ends up darker in the image because it's in shadow--so what? We're smart enough to correct for the shadow; it's not like we don't notice the shadow's there.

The distinction between "object" and "portrayal" is what Egypt Urnash was getting at with the question of language--our brains are concerned to answer the question of "what color is this object," and answer that question correctly. The "optical illusion" interpretation misconstrues "shade of grey" and "color" as applying to the portrayal, the image, rather than the portrayed, the chessboard or cube.

To put it more bluntly: if we're in a linguistic framework wherein A and B are, in fact, the same shade, i.e. looking analytically at this JPEG, the correct answer to "Are squares A and B the same color?" is "What squares are you talking about? There are only a bunch of trapezoids there." It's the same layer of abstraction--if our color-correcting mechanisms are "illusion," then so is perspective. As Wormspit reminded us, visual artists spend a lot of time learning to mimic both of these effects on two-dimensional surfaces--just like the 3D rendering software used to make these "illusions."

(Oh hell, if the point is "2D portrayals of a 3D world are illusions," I'll agree; I think we've known that for a few centuries now.)