DrawProcedural execution order with multiple cameras

I'm writing a system that uses procedurally-generated geometry to produce complex objects in the Unity scene. This works awesome with a single camera.

My scene uses multiple cameras. With two or more cameras in the scene, it seems like the result of DrawProcedural() is being executed on the GPU in the context of the NEXT camera to be drawn! This makes a bit of sense, if the DrawProcedural() call is added to the list of GPU instructions after whatever instructions end the current frame and start the next frame. I'm looking for a way to resolve this, short of hacking around to find Camera.current's index in Camera.allCameras and then feed my shader the next camera's MVP matrix, which works but introduces a one-frame delay and feels like a hack that's liable to introduce more problems later.

I call DrawProcedural like so, from a MonoBehaviour on a GameObject that has a 100% transparent object applied to a cube mesh (hacky, but makes sure OnRenderObject is called when a camera is looking at that volume of space):

 void OnRenderObject()
 {
         Matrix4x4 m = transform.localToWorldMatrix;
         Matrix4x4 v = Camera.current.worldToCameraMatrix;
         Matrix4x4 p = Camera.current.projectionMatrix;
 
         Matrix4x4 MVP = p * v * m;
 
         Graphics.ClearRandomWriteTargets();
         proceduralMaterial.SetPass(0);
         proceduralMaterial.SetMatrix("ModelViewProjection", MVP);
         proceduralMaterial.SetMatrix("model", m);
 
         Graphics.DrawProcedural(MeshTopology.Points, 640 * 480);
         Graphics.ClearRandomWriteTargets();
 }

and my geometry shader works like so:

 // Geometry Shader -----------------------------------------------------
 [maxvertexcount(4)]
 void GS_Main(point GS_INPUT p[1], uint primID : SV_PrimitiveID, inout TriangleStream<FS_INPUT> triStream)
 {
     // set up modelviewprojection matrix.
     float4x4 xf = ModelViewProjection;
 
     //snip.. v[0] is a float4 vertex.. do this 4 times to make up a quad of two tris.
     pIn.pos = mul(xf, v[0]);
     triStream.Append(pIn);
 }

All is well with a single camera, but things start to break badly as soon as two cameras are both looking at the object in question. Thoughts?