What Feynman meant in description of reversible machine and levers In chapter four of the part one of the  lectures, he mentions:

(..) A machine that we actually use can be, in a sense, almost
  reversible: that is, if it will lift the weight of three by lowering a
  weight of one, then it will also lift nearly the weight of one the
  same amount by lowering the weight of three.

and in the picture (is the fulcrum a bit of placed w.r.t the contex ?)


However, in order to get it actually to work, we must lift a little
  weight off the left pan. On the other hand, we could lift a one-unit
  weight by lowering the three-unit weight, if we cheat a little by
  lifting a little weight off the other pan.

I can't make any sense out of this. 
 A: If the example Feynman gave was "perfectly reversible", then you could move the balance up and down without doing any work. In order to overcome (even tiny amounts of) friction, you actually need to do a little bit of work. This can be done by adding or removing a little bit of weight to one of the pans of the scale.
After a cycle of moving the scale first one way, and then the other, the "state" of the scale is the same as it was before. In a perfectly reversible system, the total work done would be zero. But if you had to lift a little bit of weight to the top scale in order to tip it, then move it to the other scale to tip it back, you would have done a net amount of work on the system.
That makes this system (and any other "real world" system) irreversible.
Incidentally, the fulcrum is offset because, with 3x the weight on the left of the scale, you get "balance" when the lever on the right is 3x the lever on the left. One could have done the same example with a symmetrical scale with the same weight on each side instead.
A: Feynmann often liked to deal with realistic real world examples instead of idealised ones, to illustrate how physical principles applied in the real world. In this case I suspect that what he is getting at is that a real world hinge/balance isn't frictionless, located at a mathematical point and made of perfectly rigid materials. In the real world, you can actually balance a mass of 3 Kg against a mass of 1 Kg located three times further from the fulcrum over a small range of positions and have it be stable. To make the balance reliably tip, you need to have a slight imbalance in the weights (or distances) as due to friction and material deformation (the arms of the balance bending for example), the situation is not perfectly reversible.
