Well, I’m late to the party here, as the article linked below is now five years old; but I do have two comments.
One is that Norton is right. ALL valid thought experiments are just logical deductions of consequences of things that we have learned from experiment seem to always be true. Case in point being Einstein’s deduction of Lorentz transformations from the assumption that speed of light is invariant which is not itself deduced but extrapolated from observations (either Michelson-Morley or tests of Maxwell’s equations). In fact anyone who designs a machine is doing the same thing – thinking about logical consequenses of the assumption that previously observed patterns will continue to apply. And if the assumptions have been well tested then we strongly expect the derived laws to apply in just the same way as we expect the machine to work.
My other point is that Galileo’s beautiful argument is NOT valid. It is basically circular in that there is no reason other than prior experience to expect that two cannonballs will not fall faster when close together than when far apart. In fact it is easy to imagine scenarios where combining objects (such as water droplets) does increase their rate of fall and there is no a priori reason why all space might not be filled with a permeable but resistive medium in which the drag effect is reduced by proximity (as for cyclists in a peloton for example).