I don't understand why you say that "bytes literals require a continuation character":

>>> (b"x"
...  b"y")
b'xy'
>>> [b"x"
...  b"y"]
[b'xy']

I think the "len caching" is a misoptimization, it's useless here (most CPU time will be sent creating and wrapping the representation).
Also perhaps it would be nice to refactor things a bit, since we have both _str_parts and _bytes_parts used in exactly the same way (but that can also be done later).

As for the doc, the example would probably deserve to be a bit more "meaningful" :-)