diff -r 71876e4abce4 Include/abstract.h
--- a/Include/abstract.h	Thu Dec 15 12:40:53 2016 +0100
+++ b/Include/abstract.h	Thu Dec 15 14:43:04 2016 +0100
@@ -312,7 +312,7 @@ PyAPI_FUNC(PyObject **) _PyStack_UnpackD
    overflow. The size is also chosen to allow using the small stack for most
    function calls of the Python standard library. On 64-bit CPU, it allocates
    40 bytes on the stack. */
-#define _PY_FASTCALL_SMALL_STACK 5
+#define _PY_FASTCALL_SMALL_STACK 3
 
 /* Call the callable object 'callable' with the "fast call" calling convention:
    args is a C array for positional arguments (nargs is the number of
diff -r 71876e4abce4 Objects/abstract.c
--- a/Objects/abstract.c	Thu Dec 15 12:40:53 2016 +0100
+++ b/Objects/abstract.c	Thu Dec 15 14:43:04 2016 +0100
@@ -2701,21 +2701,12 @@ PyObject *
     return retval;
 }
 
-PyObject *
-_PyObject_VaCallFunctionObjArgs(PyObject *callable, va_list vargs)
+static Py_ssize_t
+count_vargs(va_list vargs)
 {
-    PyObject *small_stack[_PY_FASTCALL_SMALL_STACK];
-    PyObject **stack;
+    va_list countva;
     Py_ssize_t nargs;
-    PyObject *result;
-    Py_ssize_t i;
-    va_list countva;
-
-    if (callable == NULL) {
-        return null_error();
-    }
-
-    /* Count the number of arguments */
+
     va_copy(countva, vargs);
     nargs = 0;
     while (1) {
@@ -2727,9 +2718,29 @@ PyObject *
     }
     va_end(countva);
 
-    /* Copy arguments */
-    if (nargs <= (Py_ssize_t)Py_ARRAY_LENGTH(small_stack)) {
-        stack = small_stack;
+    return nargs;
+}
+
+Py_LOCAL_INLINE(PyObject *)
+object_vacall(PyObject *callable, va_list vargs)
+{
+    Py_ssize_t nargs;
+    PyObject **stack;
+    Py_ssize_t i;
+    PyObject *result;
+
+    if (callable == NULL) {
+        return null_error();
+    }
+
+    nargs = count_vargs(vargs);
+
+    if (nargs == 0) {
+        return _PyObject_CallNoArg(callable);
+    }
+
+    if (nargs <= _PY_FASTCALL_SMALL_STACK) {
+        stack = alloca(nargs * sizeof(PyObject*));
     }
     else {
         stack = PyMem_Malloc(nargs * sizeof(stack[0]));
@@ -2743,16 +2754,21 @@ PyObject *
         stack[i] = va_arg(vargs, PyObject *);
     }
 
-    /* Call the function */
     result = _PyObject_FastCall(callable, stack, nargs);
 
-    if (stack != small_stack) {
+    if (nargs > _PY_FASTCALL_SMALL_STACK) {
         PyMem_Free(stack);
     }
     return result;
 }
 
 PyObject *
+_PyObject_VaCallFunctionObjArgs(PyObject *callable, va_list vargs)
+{
+    return object_vacall(callable, vargs);
+}
+
+PyObject *
 PyObject_CallMethodObjArgs(PyObject *callable, PyObject *name, ...)
 {
     va_list vargs;
@@ -2768,7 +2784,7 @@ PyObject_CallMethodObjArgs(PyObject *cal
     }
 
     va_start(vargs, name);
-    result = _PyObject_VaCallFunctionObjArgs(callable, vargs);
+    result = object_vacall(callable, vargs);
     va_end(vargs);
 
     Py_DECREF(callable);
@@ -2792,7 +2808,7 @@ PyObject *
     }
 
     va_start(vargs, name);
-    result = _PyObject_VaCallFunctionObjArgs(callable, vargs);
+    result = object_vacall(callable, vargs);
     va_end(vargs);
 
     Py_DECREF(callable);
@@ -2806,7 +2822,7 @@ PyObject_CallFunctionObjArgs(PyObject *c
     PyObject *result;
 
     va_start(vargs, callable);
-    result = _PyObject_VaCallFunctionObjArgs(callable, vargs);
+    result = object_vacall(callable, vargs);
     va_end(vargs);
 
     return result;