Would you benchmark the performance?

How about calling Py_INCREF and Py_DECREF in PyObject_RichCompare or do_richcompare?

It is safer than checking all caller of the PyObject_RichCompare and PyObject_RichCompareBool.
And it would be faster when PyObject_RichCompareBool is called with v == w and op == Py_EQ.

    /* Quick result when objects are the same.
       Guarantees that identity implies equality. */
    if (v == w) {
        if (op == Py_EQ)
            return 1;
        else if (op == Py_NE)
            return 0;
    }