Sorry, it was incorrect microbenchmark. Correct is:

$ ./python -m timeit -r 11 -s "from collections import namedtuple as n;a = n('n', 'a b c')(1, 2, 3)" -- "a.a"
3.4          : 1000000 loops, best of 11: 0.601 usec per loop
3.5 unpatched: 1000000 loops, best of 11: 0.445 usec per loop
3.5 patched  : 1000000 loops, best of 11: 0.454 usec per loop

There is small slowdown (2%), but it is only small part of the gain of the optimization.