The solution is to add a read buffer to Unpickler (Pickler already has a
write buffer, so that why it is unaffected). I believe this would
mitigate much of the (quite large) Python function call overhead.

cPickle has a performance hack to make it uses cStringIO and PyFile 
directly (via C function calls). In Python 3, the hack was removed since
cStringIO.h and fileobject.h aren't public anymore. This explains the
0.02s deviation you are getting for dump speed.