When using GPU instancing on a small number of unique meshes, it helps to have all objects with a specific mesh on the same render queue so that GPU instancing can properly batch all objects together. Otherwise, you end up with a semi-shuffled render queue and the GPU instancing algorithm won't batch draws that aren't back-to-back. RenderQueueSo…

RenderQueueSorter.cs

public static class RenderQueueSorter
{
    private static int nextRenderQueue = 5; // Don't go over 2500! Or we leak into transparent queues
    private static Dictionary<Mesh, Material> _meshToMatMap = new Dictionary<Mesh, Material>();

    public static Material SortedMaterialFromMesh(Mesh mesh, Material sharedMaterial)
    {
        if (_meshToMatMap.ContainsKey(mesh)) return _meshToMatMap[mesh];

        return GenerateMaterialForMesh(mesh, sharedMaterial);
    }

    private static Material GenerateMaterialForMesh(Mesh mesh, Material sharedMaterial)
    {
        Material newMaterial = new Material(sharedMaterial);

        newMaterial.renderQueue = nextRenderQueue;
        nextRenderQueue += 1;

        _meshToMatMap[mesh] = newMaterial;
        return newMaterial;
    }
}

SomeMonobehavior.cs

private void Awake() {
  ...
  SetSortedMaterial();
}

private void SetSortedMaterial() {
  MeshFilter meshFilter = GetComponent<MeshFilter>();
  MeshRenderer meshRenderer = GetComponent<MeshRenderer>();

  meshRenderer.material = RenderQueueSorter.SortedMaterialFromMesh(meshFilter.sharedMesh, meshRenderer.sharedMaterial);
}