Here are my thoughts:
Suppose $F_p[x]$ is flat over $F_p[x^p]$. Then applying $- \otimes_{F_p[x^p]}F_p[x]$ to the natural injection $F_p[x^p] \to F_p[x]$ shows that the map $F_p[x] \to F_p[x] \otimes_{F_p[x^p]}F_p[x]$ given by $f \to f \otimes 1$ is injective.
So now if $a \in F_p[x] \otimes_{F_p[x^p]}F_p[x]$ is in the nilardical, then $a^p$ is also in the nilradical. But we can easily see that $a^p$ is a sum of elements of form $f(x^p) \otimes g(x^p)$ and each such sum has the form $h \otimes 1$ so then $h$ is in the nilradical of $F_p[x]$ by the above injection, so $a^p=0$.
I am not sure how to proceed from there, or even if $F_p[x]$ is actually flat over $F_p[x^p]$.
Any thoughts?
EDIT: OK, Hurkyl's idea kills it quickly and easily, but I'd still like to see if the above idea pans out.