We have our own ORM we use here, and provide strongly typed wrappers for all of our db tables. We also allow weakly typed ad-hoc SQL to be executed, but these queries still go through the same class for getting values out of a data reader.
In tweaking that class to work with Oracle, we've come across an interesting question. Is it better to use DBNull.Value, or null? Are there any benefits to using DBNull.Value? It seems more "correct" to use null, since we've separated ourselves from the DB world, but there are implications (you can't just blindly ToString() when a value is null for example) so its definitely something we need to make a conscious decision about.
I find it better to use null, instead of DB null.
The reason is because, as you said, you're separating yourself from the DB world.
It is generally good practice to check reference types to ensure they aren't null anyway. You're going to be checking for null for things other than DB data, and I find it is best to keep consistency across the system, and use null, not DBNull.
In the long run, architecturally I find it to be the better solution.