Well, yes and no.
I agree that it is annoying to have to add another parameter to get a 
returning tuple.  But you have to weigh that against adding another 
conveinence API.
On the other hand, I think the offset argument is quite convenient.  If 
we were to add a "decompress2" function, I would like to keep the 
offset because it removes the extra overhead of having to create a 
temporary buffer object, at no extra cost.

Also, note that your example code becomes more complex, for a loop.  
You would need to:
results = []
used =0
while used<len(source):
    r, u = zlib.decompress2(buffer(source, used))
    used += u
    results.append(r)

as opposed to:
    r, used = zlib.decompress2(source, used)
    results.append(r)

You have to make sure you add the buffer's hidden offset to the output 
offset.

There is a third option.  The "offset" could be a list containg the 
offset in its 0th position, a sort of "byref" passing, but that is not 
in the general Python spirit, is it?

This would give us the loop:
o = [0]
while len(source)>o[0]
    results.append(zlib.decompress(source, offset=offsetlist))