I agree that the disparity between these two values is odd. I might consider it as a feature and explain it as follows.

 
Dialog Values = get_param(gcb,'MaskValues') 

returns the values from the dialog for the mask even if they have not been “applied”

 
BlockMask = Simulink.Mask.get(gcb)
MaskValues = BlockMask.Parameters 

returns the mask parameters that are being used by the block.

For example, Matlab 7.14 (R2012a), 7.13 (R2011b), and probably earlier, has a bug in the inverse complementary error function, erfcinv, where erfcinv(eps(realmin)) returns NaN instead of a finite real value as would be expected and as indicated by the documentation (this occurs in single precision as well). I reported this bug and suggested a simple “workaround.” The bug was acknowledged and I even suggested how it might be fixed when the engineer explained the reason for the underflow. The bug was fixed in Matlab 8.0 (R2012b), but it never showed up in the online database so you’d never know that it exists/existed.

By the way, here is code that will apply a workaround in pre-R2012b versions:

x = erfcinv(y);
if verLessThan('matlab','8.0')
    if isa(p,'double')
        x(y == 2^-1074) = 27.216482834230213;
    else
        x(y == 2^-149) = 10.0198345;
    end
end

Of course verLessThan only goes back to Matlab 7.4 (R2007a), but that’s as far back as I try to retain support…

I’ve started playing around with using Mask Objects (http://www.mathworks.com/help/simulink/slref/simulink.mask.get.html) from a Simulink Mask instead of using the old reliable

 MaskValues = get_param(gcb,'MaskValues') 

, etc…
I have a dialog callback that gets that Mask Object. The problem is the Mask Object it gives you gives parameters for the Mask BEFORE it was opened, and does not include changes that you have made to the mask since opening the simulink GUI. (i.e. My dialog callback is working with old data).
My Fix: ??, go back to old reliable.