The following example seems to imply run time optimisation that I do not understand

Can anyone explain this behavior and how it may apply to more generic cases?

Example

Consider the following simple (example) functions

def y(x): # str output
    y = 1 if x else 0
    return str(y)

def _y(x): # no str
    y = 1 if x else 0
    return y

Assume I want to apply the function y upon all elements in a list

l = range(1000) # test input data

Result

A map operation will have to iterate through all elements in the list. It seems counter intuitive that breaking the function apart into a double map significantly outperforms the single map function

%timeit map(str, map(_y, l))
1000 loops, best of 3: 206 µs per loop

%timeit map(y, l)
1000 loops, best of 3: 241 µs per loop

More generically, this also applies to non standard library nested functions for example

def f(x):
    return _y(_y(x))

%timeit map(_y, map(_y, l))
1000 loops, best of 3: 235 µs per loop
%timeit map(f, l)
1000 loops, best of 3: 294 µs per loop

Is this a python overhead issue where map is compiling the low level python code where possible and consequently being throttled when it has to interpret a nested function?