The difference is in the implementation of the int type. Python 3.x uses the arbitrary-sized integer type (long in 2.x) exclusively, while in Python 2.x for values up to sys.maxint a simpler int type is used that uses a simple C long under the hood.
Once you limit your loops to long integers, Python 3.x is faster:
>>> from timeit import timeit
>>> MAX_NUM = 3*10**3
>>> def bar():
... i = MAX_NUM + sys.maxsize
... while i > sys.maxsize:
... i -= 1
...
Python 2:
>>> timeit(bar, number=10000)
5.704327821731567
Python 3:
>>> timeit(bar, number=10000)
3.7299320790334605
I used sys.maxsize as sys.maxint was dropped from Python 3, but the integer value is basically the same.
The speed difference in Python 2 is thus limited to the first (2 ** 63) - 1 integers on 64-bit, (2 ** 31) - 1 integers on 32 bit systems.
Since you cannot use the long type with xrange() on Python 2, I did not include a comparison for that function.
