protobuf.js vs JSON

While JSON is already much better than XML, it still comes with a significant overhead. Imagine you are transfering object data between two nodes: Using JSON, this might look something like {"type":"ping","time":123456789} which is, in this example, 32 bytes large, containing 2+3+3+2+1=11 bytes of bounding characters plus 4+4+4=12 bytes of string names, making up 23/32 bytes ~= 71% redundant data for each subsequent message of this type. That's much, especially if there are hundreds of clients sending thousands of messages. ProtoBuf.js, on the other hand, efficiently serializes the data to binary without losing any information, which turns out to be 7 bytes long while still being able to distinguish between two messages of different types: <0A 05 08 95 9A EF 3A>

What about Binary JSON?

BSON comes with a similar overhead. As of the specification, BSON also includes all the keys as cstrings and it also does not efficiently store integer or long values as there is no varint encoding. Take a look at the hello world example:
JSON: {"hello":"world"} (17 bytes)
BSON: <16 00 00 00 02 h e l l o 00 06 00 00 00 w o r l d 00 00> (22 bytes)

Bottom line: Not requiring a schema like in JSON/BSON is convenient but it comes at the price of size and a chance of being more error prone at least in development (no type checking, key mismatches). Additionally, Protocol Buffers are capable of storing numeric values more efficiently.

What about Protocol JSON?

PSON allows to generate an even smaller protocol than protocol buffers, at the cost of having to use it wisely.