Think of protocol buffers as serializing all your data into efficient binary data. In other words, it's not like JSON or XML, which are represented as text. Therefore, the protobufs will generally be much smaller than with other schemes. However, I don't know how CORBA encodes data, so I can't be sure protobufs will be more efficient. My GUESS is that they will, because Google is behind them, and they're used everywhere in Google's incredibly high-scaling applications.
Then let's look at ZeroMQ. ZeroMQ lists itself as being asynchronous, which means that your app should not hang. Furthermore, I don't think you'll need to break up your messages because that's what TCP/IP is supposed to do under the hood. In other words, your ZeroMQ will send a message over TCP/IP/UDP/etc., and the underlying transport layer may or may not break those messages into smaller packets.
The upshot is that while I don't have specific information about CORBA, your application, or even ZeroMQ, I'm GUESSING that yes, this application will be much more to your liking than your old approach. However, you really need to do a few experiments with real data to be sure. I would not commit to ANY new technology, even protobufs, until I was sure that they not only worked well, but worked well for my individual situation and needs.
Jeff