It is a little funny though. I expect a more common test like:

if a < b < c: pass

to be unconditionally faster than the logically equivalent:

if a < b and b < c: pass

The only difference between the two should be that b is loaded only once, which should make the chained comparison either equivalent or faster. But thanks to the additional stack manipulations needed, in microbenchmarks, this isn't the case. I was just testing a couple simple cases in ipython (3.5, Linux, x64), and the chained comparison is actually slower when short-circuiting is involved:

%%timeit -r5 a, b, c = range(3, 0, -1)
if a < b < c: pass
10000000 loops, best of 5: 33.7 ns per loop

%%timeit -r5 a, b, c = range(3, 0, -1)
if a < b and b < c: pass
10000000 loops, best of 5: 25 ns per loop

or even without short-circuiting, but with the final test failing, chained loses (by a smaller margin):

%%timeit -r5 a, b, c = 0, 1, 0
if a < b < c: pass
10000000 loops, best of 5: 42.5 ns per loop

%%timeit -r5 a, b, c = 0, 1, 0
if a < b and b < c: pass
10000000 loops, best of 5: 39.3 ns per loop

The same pattern holds if a and c are replaced with constants, e.g. if 0 < b < 2: (so only b is loaded by name). Yes, it's a small difference; when a, b, c are initialized from range(3), so all tests succeed, chained comparisons win, 44.9 ns to 54.5 ns. But it seems kind of odd that the simplest case of a chained comparison is actually a pessimization in two of the three cases.

Not saying it's worth changing (I doubt chained comparisons occur in code enough to justify the tiny savings a dedicated op code could provide), but it's not something I'd reject out of hand.