The object system isn't really fully fleshed out and I'm not sure how far forward we're going to carry it. We're looking to streamline, not further complexify, the datamodel. The original idea was to separate out the logical types defined by the object system from database types, but that doesn't mean they can't match. It just means the object system's not limited to SQL datatypes.
It also means that in theory you should be defining types that can be modelled in both PG and Oracle, so in theory you shouldn't be defining PG datatypes.
In practice, though, you're probably writing some custom thing that's not going to be generalized in a way that's going to be a standard part of the toolkit. In which case you can flush theory in the toilet.
But if there's some chance you're thinking of turning your work into something that can fit into the toolkit in a general way, then you'll want to figure out just how you'll map the PG types to Oracle.