> Usually we first merge a patch in master and after that cherry-pick it to other branches.

Right. But in this case, I needed the patch for each branch, to run the test on all branches to write a blog post :-) Here are results (blog post spoiler!):

Table showing the C stack consumption in bytes, and the difference compared to
Python 3.5 (last release before I started working on FASTCALL):

====================  ================  =====  ================  ================
Function                      2.7         3.5          3.6           3.7
====================  ================  =====  ================  ================
test_python_call      1,360 (**+352**)  1,008  1,120 (**+112**)    960 (**-48**)
test_python_getitem   1,408 (**+288**)  1,120  1,168 (**+48**)     880 (**-240**)
test_python_iterator  1,424 (**+192**)  1,232  1,200 (**-32**)   1,024 (**-208**)
Total                 4,192 (**+832**)  3,360  3,488 (**+128**)  2,864 (**-496**)
====================  ================  =====  ================  ================

Table showing the number of function calls before a stack overflow,
and the difference compared to Python 3.5:

====================  ===================  ======  ===================  ===================
Function                       2.7            3.5           3.6           3.7
====================  ===================  ======  ===================  ===================
test_python_call       6,161 (**-2,153**)   8,314   7,482 (**-832**)     8,729 (**+415**)
test_python_getitem    5,951 (**-1,531**)   7,482   7,174 (**-308**)     9,522 (**+2,040**)
test_python_iterator   5,885 (**-916**)     6,801   6,983 (**+182**)     8,184 (**+1,383**)
Total                  17,997 (**-4600**)  22,597  21,639 (**-958**)    26,435 (**+3,838**)
====================  ===================  ======  ===================  ===================