Supposing your figure was 1024 x 768 pixels, your maximum actual drawing area would be less than 1000 pixels wide, but let's call it 1000 pixels for a moment. You are packing a data range of 400 (or maybe 401, but the numbers work out nicer for 400) in to that 1000 pixels, so an x change of 1 would correspond to 2 1/2 pixels wide.
Now, if I have understood your units correctly (and I am not sure that I have), you want to plot a rectangle that is 0.02 data units wide. Multiply that by the 2 1/2 pixels per data unit, and your proposed rectangle would be 0.05 pixels wide. patch() is either going to ignore that as being too small to draw, or is going to draw it as a single pixel.
What pixel width would you need in order for such a rectagle to be two pixels wide instead of a single line? Your width would have to be at least such that W * (0.02 / 400) rounds upward to 2, so W * (0.02/400) = 1.5 at least. That requires an axes pixel width W of at least 30000, preferably higher.
Now, ordinarily you might be able to accomplish such a thing by using a vector graphics export and magnifying as needed, but the only renderer that supports vector graphics does not support transparency. The only renderer that supports transparency is OpenGL, which only supports bitmap rendering.
It has never been clear to me exactly how printing works internally and how high internal resolution can be generated for saved images. What I can suggest, though, is that your approach most likely to lead to success is to use Oliver's file Exchange contribution "export_fig".
