I figured I'd write up an answer to the general case.
Yes, the automorphism group of a parallelism always acts properly if the manifold is connected (not necessarily compact!).
Most of the steps are taken care of in Lemma 2.7 from these lecture notes by Werner Ballmann. From this we can find that if $f_n(x)$ converges for a point $x$ and $f_n\in\mathrm{Aut}(M,\omega)$ then there is a pointwise limit of $f_n$ that is smooth and an automorphism. I'll summarise the steps briefly:
Consider the constant fields $\omega^{-1}(v)$, $v\in\Bbb R^n$. These fields span the tangent space at any point, and an automorphism of $\omega$ necessarily commutes with the flow of these fields (that is $f(\varphi^t_v(x))=\varphi_v^t(f(x))$) whenever the flow is defined. By dimensional reasons these flows explore a neighbourhood of any point and by connectedness of the manifold if we start at a point $x$ we can reach any other point $y$ by iterating flows along these contant fields.
Now if $f_n\in\mathrm{Aut}(M,\omega)$ and $x$ so that $f_n(x)$ converges we have that $f_n(y)$ converges for every $y$. For let $\varphi=\varphi^{t_1}_{v_1}\circ ...\circ \varphi^{t_n}_{v_n}$ be a composition of flows taking us from $x$ to $y$, then $f_n(y)=f_n(\varphi(x))=\varphi(f_n(x))$ which converges. So we may define a function $f:M\to M$ as the pointwise limit of $f_n$.
We note that $f$ also commutes with the flows of the constant fields, which will be used to show that $f$ is smooth. For a point $x\in M$ let $U_x\subset \Bbb R^n$ open around $0$ so that the flow of constant fields at $x$ is a diffeomorphism from $U_x$ to some neighbourhood of $x$, do the same for $f(x)$ and assume that $U_x=U_{f(x)}$ by making both smaller if necessary. Then the flow of constant fields are two coordinate neighbourhoods around these points, in these coordinates the function $f$ is the identity transformation, which is smooth, hence $f$ is smooth.
Finally $f$ is an automorphism as it preserves the flow of the constant fields, meaning that $\mathcal{L}_{\omega^{-1}(v)}(f)=0$ for any constant field $\omega^{-1}(v)$, which implies that $f$ is an automorphism.
Next one can show that $f_n\to f$ uniformly on compacta. This is implies the first formulation of my question. The second formulation, that $\mathrm{Aut}(M,\omega)$ acts properly, seems a bit stronger, so I will prove that also.
To do this take a basis of $\Bbb R^n$ and as in the other answer pull back the usual inner product of this basis back to get a Riemannian metric on $M$, denote with $d:M\times M\to \Bbb R$ the induced distance function. Automorphisms of $\omega$ automatically preserve this metric and thus also the distance function.
Now we can appeal to two simple general results:
If $(X,d)$ is a locally compact metric space and $f_n$ a sequence of isometries that converge pointwise to an isometry $f$ then there is a sub-sequence that converges uniformly on compacta.
Let a group $G$ act on metric space $(X,d)$ by isometries, then (if $f_n\in G$, $x\in X$ so that $f_n(x)$ converges $\implies$ $\exists$ a subsequence $f_{n_k}$ and an $f\in G$ so that $f_{n_k}\to f$) is equivalent to ($G$ acts properly on $X$).
The first step follows from considering $\sup_{z\in K}d(f_n(z), f(z))$ for a compactum $K$. If that doesn't converge to $0$ there must be a sequence $z_n\in K$ so that the above is $>\epsilon$, by compactness of $K$ we may assume $z_n$ converges to some $z\in K$. Then
$$\epsilon<d(f_n(z_n),f(z_n))≤ d(f_n(z_n),f_n(z)) + d(f_n(z),f(z)) + d(f(z), f(z_n)).$$
However since $f$ and $f_n$ are isometries the first and the last summand may be replaced by $d(z_n, z)$ which converges to $0$. The middle summand also converge s to $0$ since $f_n$ converges pointwise to $f$. This contradiction means that $f_n$ has a sub-sequence that converges uniformly on $K$. By doing a diagonalisation argument and making use of an exhausting sequence of compacta (use second-countability of metric spaces and local compactness), we find a sub-sequence $f_{n_k}$ convering uniformly on any compactum.
Next, acting properly means that whenever $f_n(x_n)$ converges with $x_n$ convergent it follows that $f_n$ admits a convergent sub-sequence. Thus for the equivalence desired the direction "$\impliedby$" is trivial and we show "$\implies$".
Now suppose the left-hand side of the equivalence holds and we have sequences $f_n(x_n)\to y, x_n\to x$, then:
$$d(f_n^{-1}(y),x) ≤ d(f_n^{-1}(y),x_n)+ d(x_n,x) = d(y, f_n(x_n)) + d(x_n,x)$$
hence $f_n^{-1}(y)$ converges to $x$. The left-hand side implies that $f_n^{-1}$ then converges to some $f^{-1}$, thus $f_n$ converges to $(f^{-1})^{-1})$.