Hm, Qt's internal processing suggests it might actually be intentional?

void QMdiAreaPrivate::_q_processWindowStateChanged(Qt::WindowStates oldState,
                                                   Qt::WindowStates newState)
{
    if (ignoreWindowStateChange)
        return;

    Q_Q(QMdiArea);
    QMdiSubWindow *child = qobject_cast<QMdiSubWindow *>(q->sender());
    if (!child)
        return;

    // windowActivated
    if (!(oldState & Qt::WindowActive) && (newState & Qt::WindowActive))
        emitWindowActivated(child);
    // windowDeactivated
    else if ((oldState & Qt::WindowActive) && !(newState & Qt::WindowActive))
        resetActiveWindow(child);

    // windowMinimized
    if (!(oldState & Qt::WindowMinimized) && (newState & Qt::WindowMinimized)) {
        isSubWindowsTiled = false;
        arrangeMinimizedSubWindows();
    // windowMaximized
    } else if (!(oldState & Qt::WindowMaximized) && (newState & Qt::WindowMaximized)) {
        internalRaise(child);
    // windowRestored
    } else if (!(newState & (Qt::WindowMaximized | Qt::WindowMinimized))) {
        internalRaise(child);
        if (oldState & Qt::WindowMinimized)
            arrangeMinimizedSubWindows();
    }
}