Consider the following:

import decimal

class MyDecimal(decimal.Decimal):
    def __new__(cls, value):
        return super().__new__(cls, value)

a = decimal.Decimal('1.0')
b = MyDecimal(a)
c = MyDecimal('1.0')

print(type(a), type(b), isinstance(b, MyDecimal), type(c), isinstance(c, MyDecimal))

Running the above in 3.3 produces:

<class 'decimal.Decimal'> <class 'decimal.Decimal'> False <class '__main__.MyDecimal'> True

Which shows that Decimal.__new__(cls, Decimal()) will always return its argument regardless of cls.

New changeset 1b6c972457e6 by Stefan Krah in branch '3.3':
Issue #16431: Use the type information when constructing a Decimal subtype
http://hg.python.org/cpython/rev/1b6c972457e6

New changeset 7fcc58811e82 by Stefan Krah in branch '3.3':
Issue #16431: Also fix the opposite direction.
http://hg.python.org/cpython/rev/7fcc58811e82

"Py_INCREF(v); return v;" should be used only for immutable types, not for subclasses. in 3.2, the code below prints "3, None":

import decimal
class MyDecimal(decimal.Decimal):
    x = None
    def __new__(cls, value):
        return super().__new__(cls, value)
        return obj
a = MyDecimal("1.1")
a.x = 3
b = MyDecimal(a)
print(a.x, b.x)

New changeset 9a701e8ec2c9 by Stefan Krah in branch '3.3':
Issue #16431: Finally, consider all permutations.
http://hg.python.org/cpython/rev/9a701e8ec2c9

Thanks, Amaury. Should be fixed now.