why are physics engines convincingly fake when we have all the equations and the real physics data that we could input to create a real-life physics engine?
Because nobody has the computer power to simulate even an atom using the actual, real laws of physics. There are people using months of running time on supercomputers to approximately simulate single protons. Yes, we know the physics, but it's just insanely frackin' intractable.
Ok, so maybe you don't want to do things down to the level of quantum chromodynamics. You never got all that quantum stuff, anyway. Plain Newton and Maxwell is good enough for a game! Very good. We'll put in Newton and Maxwell, and arbitrarily designate atoms as the, um, atomic minimal objects of our physics sim. How many do we need? Well, in a gram of hydrogen there are Avogadro's number of hydrogen atoms; 6.02 x 10^23. The number of interactions per time-tick is, of course, O(n^2) in the number of atoms. So, if we optimistically assume that an interaction can be calculated in one floating-point operation, a computer would need to run at (*)... 181 GGGGGHz to do one frame per second. For one gram of hydrogen.
* Footnote: A GGHz is a billion GHz; a GGGHz is a billion GGHz... Yes, I'm aware I'm somewhat simplifying the conversion between FLOPS and clock speed; bear with me, it doesn't affect the argument.