Answering to the question of why the OP method gives a lower bound for the codomain, consider (e.g.) the following:
$$
\begin{gathered}
\left( {z^{\,2} + 1} \right) \geqslant 2z\quad \left( {y^{\,2} + 1} \right) \geqslant 2y\quad \Rightarrow \quad \left( {z^{\,2} + 1} \right)\left( {y^{\,2} + 1} \right) \geqslant 4\,z\,y\quad \Rightarrow \hfill \\
\Rightarrow \quad \left( {z^{\,2} + 1} \right)\left( {y^{\,2} + 1} \right) = 4\quad \left| {\;\left| z \right| = 1\; \wedge \;\left| y \right| = 1} \right. \hfill \\
\end{gathered}
$$
Here, since each single inequality bears the $ \geqslant$ sign, implies that the equality is attained for some values of the variable.
When multiplied together , keeping the variables distinct , then surely for all
the $(y,z)$ combining the above values, the equality will be attained.
However, when in it we place $(y(x), z(x))$ we are going to restrict the values of
$y$ and $z$ to stay on the curve parametrized in $x$, which might not allow
$y$ and $z$ to attain "contemporarily" the equality (allow me to say that you might have $( = )\, \times \, ( > )$).
In particular, the substitution
$$
\left\{ \begin{gathered}
y = 1/\sin ^{\,2} x \hfill \\
z = 1/\cos ^{\,2} x \hfill \\
\end{gathered} \right.
$$
will produce the situation which is well evidentiated in the picture.

the equality occurs only if $a=b$
which is not possible here for obvious reason.
– lab bhattacharjee Oct 26 '16 at 09:54