Try this:
x = linspace(0, 6, 500);
upper_curve = sqrt(x);
lower_curve = x.^2 - 3;
idx = (upper_curve >= lower_curve) & (upper_curve >= 0);
figure(1)
plot(x, upper_curve, '-k', x, lower_curve, '-k')
hold on
patch([x(idx) fliplr(x(idx))], [upper_curve(idx) fliplr(max([lower_curve(idx); zeros(1,nnz(idx))]))], 'g')
hold off
axis([-1 6 -6 6])
grid
The logic and approach are the same as with your previous Question, with the addition of the logic defining the lower border. That is the maximum of either zero or the lower curve. That begins as a (2xN) matrix of a (1xN) vector of zeros and a (1xN) vector describing the lower curve. The max function creates a row vector of the maximum of each column, producing a vector that is the maximium of zero or the value of the lower curve at each ‘x’ value (defined in the first argument to the patch function here).