Ok, here's what I did:

>>> import socket
>>> s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
>>> s.bind(('', 9009))
>>> s.listen(5)
>>> s.accept()

Now, I opened a second Python interpreter in which I
typed this:

>>> import socket
>>> s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
>>> s.connect(('localhost', 9009))

In the first interpreter I did this:

>>> s.accept()
(<socket._socketobject object at 0x40163e14>,
('127.0.0.1', 33059))
>>> s1 = _[0]
>>> s1.settimeout(3)
>>> fd = s1.makefile()

Then I tested that the timeout worked correctly.  Still
in the first interpreter:

>>> fd.readline()
Traceback (most recent call last):
  File "<stdin>", line 1, in ?
  File "/usr/local/lib/python2.3/socket.py", line 338,
in readline
    data = self._sock.recv(self._rbufsize)
socket.timeout: timed out
>>> fd.readline()

Now, while that was blocking, I did this in the second
interpreter:

>>> s.send('foo')
3

Which caused this in the first interpreter (as
expected, since I didn't send a newline):

Traceback (most recent call last):
  File "<stdin>", line 1, in ?
  File "/usr/local/lib/python2.3/socket.py", line 338,
in readline
    data = self._sock.recv(self._rbufsize)
socket.timeout: timed out
>>> fd.readline()

While that was blocking, I did this in the second
interpreter:

>>> s.send('bar\n')
4

Finally sending a newline.  But lo and behold!  In the
first interpreter I got this:

>>> fd.readline()
'bar\n'

Alas, my 'foo' has been lost!

Anyway, the documentation does explicitly state that
the socket should be in blocking mode, *implying* that
it does no buffering, but it doesn't say anything about
timeouts.  Ideally, the file object would buffer enough
data until the readline could return meaningfully, but
failing that, the documentation should probably be
updated to mention that timeouts shouldn't be used with
readline on the returned file object.