I hadn't heard of protobufs before now. But if I understand it correctly, I don't think it addresses (what I think is) the fundamental problem with static typing in a distributed system.
It's that the your static types that exist at development/compile time aren't really strongly related to the runtime data. Meaning, the strong guarantees that static types make aren't actually true.
Systems deal with this by recreating the relationship: they convert incoming data to instances of the static types. But you have to note, this code exists to satisfy the static type system. If you didn't have a static type system, you don't need that code: you don't have to write it, test it, maintain it, etc.
protobufs looks like one way to recreate the relationship between your runtime data and the static types. It looks pretty nice, but it has its own complexities -- another language to learn on top of anything else you're doing, another tool to install and learn with its own quirks (I notice the generated C++ code files have a .cc extension), a code generation step to integrate into your build, new versions of protobuf to be integrated to get new features). Stuff you don't need to deal with if you don't have a static type system.
But it's not just that its wasteful. It also encourages you to deal with changes to your data model early, in a generic way (because your inflexible static type system pervades the entire code base... you can't do anything with the data until you wrestle it into an instance of a static type). But that's often not the best place to do it because at that point, the system doesn't know what will be done with the data.
Not that static types don't have their place. They certainly have their uses, but I think it's in small, inner-boxes, not throughout a codebase. E.g., let's make it something you have to opt-in to on a case-by-case basis, not a pervasive assumed default. Sure, provide a mechanism for type hinting to drive helpful static code analysis, like auto-complete and refactoring tools.
Perhaps provide some kind of flexible dynamic-type/static-type bridge that helps you solve dynamic-type/static-type issues rather than encourage them. E.g., something that at run time keeps track of the difference between runtime data and the static type it is supposed to be according to the type hinting.