What is the performance impact of hundreds of jointed RigidBodies?

I recently started making a game where the point is to build structures made of blocks and other parts and the built structures can move around and collide with others, kind of like the game Scrap Mechanic.

The blocks have to be joined together (FixedJoint was too weak, so I had to use a ConfigurableJoint), but the problem might be that after a while, hundreds or thousands of objects will be joined together, all with their own RigidBody, Colliders and ConfigurableJoints.

So I’m asking this:

Do many RigidBodies impact the performance too much?

Would it be better to just child the objects to one GameObject with a RigidBody?

Or said otherwise:

Are Compound Colliders noticeably better than Jointed RigidBodies?

Specific example: Imagine a statue/car/whatever structure made of hundereds of small 1x1x1 blocks, occasionally a special part, like a piston. All blocks have their own non-kinematic, gravity-enabled RigidBody, at least one non-trigger Collider (and several trigger ones), and each block would be joined to its neighbours with a ConfigurableJoint with Fixed movement and rotation.

Would it be better to leave it this way, or should I make an extra GameObject with a RigidBody to hold all these objects as children without their own RigidBody and with no Joints, essentially creating a compound collider?

I didn’t find a satisfactory answer online yet. I know that premature optimization is bad, but deciding about this early would save a lot of changing afterwards. When I realize that the game is slow, it might be too late to simply rewrite the thing.

Unless someone has experience, the thing to do is create a test case where you create a scene with one system and try things out on it, outside of the game. Complex physics are costly and you will need to find hacks to limit the computation workload. There’s no way you want hundreds of blocks all connected with joints to its neighbours. You’d want all stationary objects to have no joints, and to connect objects up as needed, I think.

Don’t think of it as premature optimisation, but correct solution finding. If you can’t find a way to do what you want effectively, all those ideas are worthless. Spend a week experimenting to see if your game idea is a flier on the platforms you’re targeting.