This is a neat question. I'm not sure if you tried this or not, but a first step might have been to test out the stl to make sure it displays a Penrose Triangle (it does, I downloaded a free STL viewer and it looked fine).
I think that you are pretty close to having the correct code (though the first input to patch, fv, is invalid, or at least I couldn't get it to work correctly). Details on the patch function can be found here http://www.mathworks.com/help/matlab/ref/patch.html. Note that for 3D images, it states
patch(X,Y,Z,C) creates a patch in 3-D coordinates. If the coordinate data does not define closed polygons, patch closes the polygons . . . a patch object is one or more polygons defined by the coordinates of its vertices.
So four inputs are required for the 3D image - (paraphrasing from the link) the elements of X,Y and Z specify the vertices of a polygon. If X,Y and Z are m-by-n matrices, MATLAB draws n polygons with m vertices. C determines the color of the patch.
The import_stl_fast function has a second mode (2) that returns the vertices (and triangular norms) from the ASCII STL file, so if we do that (instead of getting the points, triangles and norms) it may be a little easier to plot the Penrose Triangle
[vertices,norms] = import_stl_fast(file,2);
The trick is to take vertices and put it in a format that is applicable to patch. From above, if the X,Y, and Z matrices are mxn, then n polygons with m vertices will be drawn. Since we are drawing triangles (as read in from the STL file) we need to initialize the matrices to be 3xn as each of our polygons have three vertices
n = length(vertices)/3;
X = reshape(vertices(:,1),3,n);
Y = reshape(vertices(:,2),3,n);
Z = reshape(vertices(:,3),3,n);
C = zeros(3,n);
Then you can call patch similar to what you did
patch(X,Y,Z,C,'FaceColor', [0.8 0.8 1.0], ...
'EdgeColor', 'none', ...
'FaceLighting', 'gouraud', ...
Try the above and see what happens!
EDIT a note on use of reshape: the vertices matrix has three columns for each of the x,y and z coordinates of a vertex. We take each column that is a 20157x1 vector and re-shape it (without any loss of data) into a 3x6719 matrix.