Can someone give me a reference for the following or an idea on why it is true? (This is taken from remark 1.5 on page 5 of http://arxiv.org/abs/0810.0794.)
Suppose we have an algebraic group $G$ acting on an algebraic variety $X$. If $G$ is a connected unipotent group over $k$ ($\operatorname{char} k > 0$), one can show that the forgetful functor from the equivariant derived category of $\ell$-adic sheaves on $X$ to the derived category of $\ell$-adic sheaves is fully faithful.
While the statement is a tautology if one uses Definition 1.3 in the cited paper (the "naive" equivariant derived category), I took the question to be about the usual definition of equivariant derived category (e.g. as in Bernstein-Lunts); as such, it has content.
The proof that the two definitions agree is in Appendix C of the same paper. In particular, this proof shows that the forgetful functor from the usual equivariant derived category is fully faithful. Below is a sketch of a proof of fully faithfullness in case it helps.
To my mind, the key point is that a unipotent group $G$ is isomorphic as a variety over $k$ to an affine space. In particular it has trivial cohomology: $H^j(G;\mathbb Q_\ell)=0$ for $j>0$. The forgetful functor
$$ F:D_G(X) \to D(X)$$ has a right adjoint, $E$. To show that $F$ is fully faithful, we just need to check that the unit $$1_{D_G(X)} \to EF$$ is an isomorphism (this is easy to see).
In the language of stacks, this isomorphism is fairly intuitive: there is a morphism $f:X \to X/G$ which is a $G$-torsor - in partiuclar, it is a "fibre bundle with contractible fibres". In this language, the functors $E$ and $F$ are $Rf_\ast$ and $f^\ast$ respectively. We have $$Rf_\ast f^\ast (\mathcal F) = \mathcal F \otimes Rf_\ast(\mathbb Q_{\ell X})$$ by the projection formula. As the fibres of $f$ are contractible $Rf_\ast (\mathbb Q_{\ell X}) \simeq \mathbb Q_{\ell X/G}$. Thus $Rf_\ast f^\ast \mathcal F \simeq \mathcal F$ as required.
