This answer is basically a summary of the comments above. First, for $N$ equals $1$ and for $d\geq 2$, for every integer $n$, the morphism is not birational. One can analyze the pushforward of the structure sheaf; there is something about this in "The Kodaira dimension of spaces of rational curves on low degree hypersurfaces".

For $N$ equals $2$, denote by $\mathbb{P}H^0(\mathbb{P}^2,\mathcal{O}(d))$ the projective space parameterizing degree $d$, effective Cartier divisors in $\mathbb{P}^2$. Denote by $\Sigma^o(d)$ the locally closed subset of $\mathbb{P}H^0(\mathbb{P}^2,\mathcal{O}(d))$ parameterizing Cartier divisors that are reduced, irreducible, with precisely $\delta = (d-1)(d-2)/2$ ordinary double points and no other singularities, i.e., the Severi variety of degree $d$, genus $0$ curves in $\mathbb{P}^2$. Denote by $\Sigma(d)$ the Zariski closure of $\Sigma^o(d)$ in $\mathbb{P}H^0(\mathbb{P}^2,\mathcal{O}(d))$. Then there is a "Hilbert-Chow morphism",
$$ FC:\overline{\mathcal{M}}_{0,0}(\mathbb{P}^2,d) \to \Sigma(d).$$
This morphism "factors" the evaluation morphisms above. This morphism is birational, and thus also the evaluation morphisms are birational for $n\gg 0$. Via linear projections, it follows that for every $N\geq 2$, for every $d\geq 1$, for every $n\gg 0$, the evaluation morphism is birational.

However, $\Sigma(d)$ is not normal, and thus also the images of the evaluation morphisms will not be normal when $N$ equals $2$, when $d\geq 3$, and when $n\gg 0$. Consider, for instance, the irreducible closed subset $D_{1,d-1}$ of $\Sigma(d)$ whose general points parameterize reducible Cartier divisors, one component of which is a line $L$, and the other component is a Cartier divisor $D$ of degree $d-1$ parameterized by $\Sigma(d-1)$. The issue is that $L\cap D$ consists of $d-1$ points. For each point, we can deform just that node of $L\cup D$ to obtain a general point of $\Sigma(d)$. Thus, $\Sigma(d)$ has $d-1$ branches along $D_{1,d-1}$. Therefore $\Sigma(d)$ is not normal.

Actually, now I see that this problem persists for $N\geq 3$. Consider curves in $\mathbb{P}^N$ of the form $L\cup D$, where $D$ is a general curve of genus $0$ and degree $d-1$ in $\mathbb{P}^N$, and where $L$ is a line that intersects $D$ in $2$ points. Locally near such a point, the image $\Sigma(d)$ (not a closed subset of the Chow variety) has two branches. Since $\Sigma(d)$ is not normal, also the images of the evaluation morphisms are not normal.

probablynormal. $\endgroup$ – Jason Starr May 30 '15 at 10:32