Using the TikZ spy library to spy on a pixel value in beamer

Question

I tried to use the TikZlibrary spy to zoom in on a certain pixel image; I am drawing in the image using a scope and red about the problem concerning coordinates in scopes here Magnify part of picture with TikZ spy, expressed in relative scoped coordinates

My problem is, that spy seems to place the point in the image correctly but then chooses the wrong part to magnify, for example in the MWE

\documentclass[german,10pt,xcolor=colortbl,compress]{beamer}
\usepackage{xunicode}
\usepackage{calc}
\usepackage{graphicx,tikz}
\usetikzlibrary{spy,calc}
\RequirePackage[no-math]{fontspec} 
\RequirePackage{xltxtra}
\setbeamertemplate{navigation symbols}{}
\begin{document}
    \begin{frame}{Example}{Spy on Images in Beamer}
        \begin{tikzpicture}[spy using outlines={circle,size=.1\textwidth, magnification=3, connect spies}]
            \node[anchor=south west,inner sep=0pt] (image) at (0,0) {%
            % \includegraphics[height=.5\textwdth]{exampleA}
            \begin{tikzpicture}
                \draw[draw=blue,fill=green!50] (0,0) rectangle (.5\textwidth,.5\textwidth);
                \shade[top color=black, bottom color=white] (.25\textwidth,.25\textwidth) rectangle (.5\textwidth,.5\textwidth);                    
            \end{tikzpicture}
            };
            \begin{scope}[x={(image.south east)},y={(image.north west)}]
                % Draw besides image
                \shade[top color=red, bottom color=blue] (1.025,.66) rectangle (1.075,1);
                \node (spy1) at (128/256,192/256) {};
                \coordinate (spyto1) at (-.5,.75); 
                \spy [black,thick] on (spy1) in node at (spyto1);
                \node (spy2) at (256/256,224/256) {};
                \coordinate (spyto2) at (-.5,.25); 
                \spy [black,thick] on (spy2) in node at (spyto2);
            \end{scope}
            % Adress a point in “pixels”
        \end{tikzpicture}
\end{frame}
\end{document}

The magnification should be directly at the border of the green part to the shade. nevertheless, the magnification shows only green and playing around with the second one, i think it's off by both directions to the lower left.

I have seen no way to obtain information about the magnification process, so - any clues, which computation is still affected by some wrong scoped coordinates would be very helpful.

Edit: As Andrew noted, this is fixed (also for me now) with TikZ/PGF 3.0; though if anybody could present details on what is going wrong here - feel free to share.