I've performed a little benchmark comparing the performances of Cerealizer, cPickle, Jelly + Banana and Jelly + cBanana (from Twisted Matrix) for serializing a list of 2000 objects defined as following :

class O:
  def __init__(self):
    self.x = 1
    self.s = "jiba"
    self.o = None

with self.o referring another 0 object in the list.

The results are (with Psyco) :

With old-style classes:
	cerealizer
	dumps in 0.0619530677795 s, 114914 bytes length
	loads in 0.0313038825989 s


	cPickle
	dumps in 0.0301840305328 s, 116356 bytes length
	loads in 0.023097038269 s


	jelly + banana
	dumps in 0.168012142181 s 169729 bytes length
	loads in 1.82081913948 s


	jelly + cBanana
	dumps in 0.082946062088 s 169729 bytes length
	loads in 0.156159877777 s


With new-style classes:
	cerealizer
	dumps in 0.0575239658356 s, 114914 bytes length
	loads in 0.028165102005 s


	cPickle
	dumps in 0.07634806633 s, 116428 bytes length
	loads in 0.0278959274292 s


	jelly + banana
	dumps in 0.156242132187 s 169729 bytes length
	(TypeError; I didn't investigate this problem yet although it is surely solvable)


	jelly + cBanana
	dumps in 0.10772895813 s 169729 bytes length
	(TypeError; I didn't investigate this problem yet although it is surely solvable)

feature	pickle	cPickle	Cerealizer	Jelly + Banana	Jelly + cBanana
Language	Python	C	Python	Python	Python + C
Speed	Very slow	Very fast	Fast; can even beat cPickle on new-style classes	Very slow	Slow
Secure	No		Yes	Yes
File length	Medium		Medium, roughly equivalent to pickle	Medium, files seem slighly more bigger on my bench
Human-readability of files	Poor		Very poor	Good
Supports C-defined types	Yes		Yes	Yes, although see (*)

(*) Jelly handles them, but tuples and objects in a cycle are first created as _Tuple or _Dereference objects; this works for Python classes, but not with C-defined types which expects a precise type (e.g. tuple and not _Tuple), due to static typing.