It would be hard to get the same optimization in the peepholer.

For the first example, "if not a and b:", it is easy, just replace UNARY_NOT+POP_JUMP_IF_FALSE with POP_JUMP_IF_TRUE. The more complex third example, "if not (a and b) and c:", would need multiple passes (and more complex examples can need more passes, having the quadratic complexity in corner cases). The second example, with chained comparisons, is the most complex. It uses the fact that after the conditional jump and some stack operations we got a false value on the stack. This is too complex for the peepholer.