While the other existing answers give simple algebraic reasons for this fact, it is actually far more useful in general to see this fact as a special case of the smoothing technique. In particular, for any concave function $f$ on domain $D⊆ℝ$, we have that $f(a+b) ≥ f(a'+b')$ for every $a,b,a',b'$ such that $a+b = a'+b'$ and $a' ≤ a,b ≤ b'$. That is, pushing the points $a,b$ apart while preserving their sum decreases the total value of $f$ on them. In your case you simply have $(a,b) = (z,x-z)$ and $(a',b') = (0,x)$ and $f$ being the real-square-root function.
This general smoothing technique is extremely powerful if you know how to use it. For example it gives a one-line proof of AM-GM inequality, and similarly a short proof of Jensen's inequality. In discrete mathematics it is sometimes called a swapping argument (here is an example usage). In real analysis, it can be used in conjunction with a compactness argument to prove theorems that can be quite difficult to prove without (such as the two continuous optimization theorems in this post).