Yes, you are correct. It does give the correlation coefficients for the every lag values. By using 'coeff' the following changes are made:-
xcorr(x,y,'coeff') = xcorr(x,y)/(sqrt(sum(abs(x).^2)*sum(abs(y).^2)))
Although, this xcorr(x,y,'coeff') is different from the pearson's definition of corr. coeff. in the sense that it doesn't subtract mean of x and y, your application will give the correct ans as the zscore already performs the normalization.
Check out the documentation for xcorr if you haven't already.
Hope this helps!