I just changed the way the software is determining the upper character of a keyboard key (the one that you can obtain with the Shift key) – you can download the beta2 from here. The previous solution was particularly bad because the software would actually generate keystrokes when it started to determine upper/lower character pairs.
I was warned that this might trigger accelerator keystrokes on some machines when the software starts. It should not happen any more. If you still find something wrong with the new solution, please report it to me.
(As you might know, when you long-press a keyboard key in Math-o-mir, the key turns uppercase. It is for this feature that the software needed to determine upper/lower character pairs.)