I agree. I've written many home-grown ORMs as well as used a few well-known ones. It's definitely easier in some ways writing your own because you understand all of the details & it has nothing in it that is unneeded. The downside is that you don't have the community support maintaining the code and providing support resources. On the flip-side I've spent days researching and tweaking hibernate mapping files, occasionally making me want to throw my laptop out the window!
One thing is for sure, no matter what ORM or similar solution you use, if you don't know how to observe and understand the SQL that is generated then you won't be able to use it efficiently.
If there were a way to scale an ORM so it can modularly add features/complexity as you need it (ie, customize the metamodel complexity), that would be the most ideal approach.
On the other hand, testing thousands to millions of possible use cases in combined codebases is a nightmare.