this is what i have finally stuck with:

def_struct_primitive(meshop_getAllElements, meshop, "getAllElements");
Value* meshop_getAllElements_cf(Value** arg_list, int count)
{
	check_arg_count_with_keys(getAllElements, 1, count);
	Mesh* mesh;
	BOOL deleteIt = FALSE;
	TriObject *pTri;

	if (is_node(arg_list[0]))
	{
		INode* node = arg_list[0]->to_node();
		TimeValue t = MAXScript_time();
		pTri = GetTriObjectFromNode(node, t, deleteIt);
		if (pTri)
		{
			mesh = &pTri->GetMesh(); 
		}
	}
	else if (is_mesh(arg_list[0])) 
	{
		mesh = arg_list[0]->to_mesh();
	}

	if (mesh)
	{
		int level = is_number(key_arg(level)) ? key_arg(level)->to_int() : (key_arg(level) == n_verts) ? 1 : (key_arg(level) == n_faces) ? 3 : 3;

		int numf = mesh->numFaces;
		BitArray faces(numf);
		faces.SetAll();
		int numv = mesh->numVerts; 
		Tab<Tab<int>> verts;
		verts.SetCount(numv);

 		Array* elements = new Array(0);  

		for (int i = 0; i < numf; i++)
		{
			Face& f = mesh->faces[i]; 
			for (int k = 0; k < 3; k++) verts[f.v[k]].Append(1, &i);
		}

 		for (int i = 0; i < numf; i++) if (faces[i])
		{
			Tab<int> element;
			element.Append(1, &i);
 			BitArray fbits(numf);
			BitArray vbits(numv);

			for (int j = 0; j < element.Count(); j++) 
			{
				int fi = element[j];
				if (!fbits[fi])
				{
					Face& f = mesh->faces[fi]; 
					for (int k = 0; k < 3; k++) 
					{
						int v = f.v[k];
						if (vbits[v]) continue;

						element.Append(verts[v].Count(), verts[v].Addr(0));
						vbits.Set(v);
					}
					fbits.Set(fi);
					faces.Clear(fi);
				}
			}
			if (level == 3) elements->append(new BitArrayValue(fbits));
			else if (level == 1) elements->append(new BitArrayValue(vbits));
		}
		return elements;
	}
	return &undefined;
}

it gives me an option to get elements of verticies for almost free.

all memory is used by data for return to mxs anyway

there is no leak in system memory

with slight modification this method might be changed do return elements for specified faces or only one face