IPLD is an abstract data model. It basically says "all Merkle DAGs have basic properties in common, so let's use those common properties to interlink DAGs across different systems." This allows you to treat the entire decentralized web -- and all DAGs anywhere -- as one giant inter-linkable information space. This makes any system that uses IPLD really powerful and dynamic, opening lots of paths for interoperability.
If a system incorporates IPLD into its code, it can link to other hash-linked data in any system. So if two systems both use IPLD they will be able to bidirectionally reference each other's data. If only one system uses IPLD and the other does not (ie. IPFS uses IPLD but git doesn't) then your IPLD-based system (IPFS) can point into the other system's data (git) but that other system won't be able to point back.
The benefits of incorporating IPLD into your code base are huge, and the cost/overhead is minimal. That makes me suspect that we will see all hash-linked systems gradually incorporate IPLD, which will make the decentralized web exponentially more powerful and flexible.