Are mesh colliders much slower than box colliders even when there are no collisions?

I’ve got an environment off the asset store with hundreds of mesh models, and the developer has been quite conservative with placing box colliders only where necessary, with no mesh colliders. Trouble is I need very precise physics, in my game it’s extremely obvious when the player has collided with an invisible wall.

I also need to keep a very close eye on performance, so am I looking at a serious performance hit if I convert everything to mesh colliders? The key thing is there are almost no collisions in normal gameplay, and the only dynamic object is the player, which uses a box collider.

Here’s the item: Unity Asset Store - The Best Assets for Game Making


Not really an answer, but more than a comment:

Typical collision code does a lot of “culling” of colliders, before it even gets to the specific type. I think, after the OcTree “things I’m anywhere near” check, people check Axis-Aligned-Bounding-Boxes. Then, only if those overlap, will it check the actual colliders.

So, in theory, only long diagonal or L-shaped (huge bounding boxes) mesh colliders will be a problem. Obviously, run a quick test with 1,000 mesh colliders randomly scattered. But I’m pretty sure a MeshCollider over in the far corner is the exact same cycles as a BoxCollider.