Protocol buffer was created to be smaller, faster, and simpler, and must provide an automated mechanism to serialize structured data. You decide the structure, then you can use the generated source code to read and write data to various streams and to many languages. You can change your data structure without breaking the old format.
As compared to XML serialization, it is:
Simpler.
Generates DAO classes, which works well for the programmers.
Footprint is 3 to 10 times smaller.
It is 20 to 100 times faster, which is a big benefit.
Much simpler, thus less ambiguous.
Efficiency: Protocol buffers are generally much more efficient at transmitting the same amount of data than Java binary serialization.
Portability: Java binary serialization is not widely implemented outside Java, as far as I'm aware (unsurprisingly). Robustness in the face of unrelated changes. Unless you manually specify the serializable UUID, you can end up making breaking changes without touching data...