gprof (--enable-profiler) results:

Each sample counts as 0.01 seconds.
  %   cumulative   self              self     total           
 time   seconds   seconds    calls  ms/call  ms/call  name    
 18.18      0.16     0.16  2011055     0.00     0.00  vgetargs1
  7.95      0.23     0.07  2017033     0.00     0.00 
PyObject_GenericGetAttr
  6.82      0.29     0.06  2002006     0.00     0.00 
_BufferedReader_read_unlocked
  5.68      0.34     0.05  2012123     0.00     0.00  convertsimple
  4.55      0.38     0.04  2013284     0.00     0.00  PyCFunction_Call
  4.55      0.42     0.04  2004543     0.00     0.00  PyLong_AsSsize_t
  3.41      0.45     0.03  2004565     0.00     0.00  PyNumber_Index
  3.41      0.48     0.03  2002177     0.00     0.00  float_argument_error
  3.41      0.51     0.03  1000497     0.00     0.00 
PyLong_AsLongAndOverflow
  2.27      0.53     0.02  2012123     0.00     0.00  convertitem

It looks like PyArg_ParseTuple() should be optimized ;-)