We will first prove that coverings can be found if n=3k and if n=3k−1. When k=1 these are the cases n=2 and n=3 for which the diagrams below show a covering by trexes. In these diagrams, the unit hexagons are represented by their centres.

To extend this for all k≥1, we observe that, for any n≥1, the region obtained by removing an n-level honeycomb from an (n+3)-level honeycomb can be covered by trexes without overlap, see left diagram below. The statement follows now from the cases n=2 and n=3.
Alternatively, we may cover the large hexagon formed by the centres of the unit hexagons by 3 congruent parallelograms, each covering n(n+1) centres, as shown in the diagram on the right below.

When the number of hexagon centres along one side of such a parallelogram is divisible by 3, we can cover it by trexes. More specifically:
* if n=3k then each parallelogram of size 3k(3k+1) can be covered by k(3k+1) trexes;
* if n=3k+2 then each parallelogram of size (3k+2)(3k+3) can be covered by (k+1)(3k+2) trexes.

Next we prove that no coverings exist if n=3k+1. We present three different proofs. In all of them we label the centres of the hexagons by 0, 1, 2 with 1 in the centre of the honeycomb and so that two hexagons with the same label never share an edge. It follows that every trex contains each label 0, 1, 2 exactly once, hence the honeycomb (with central hexagon removed) can only be covered with trexes when it contains the same number of 0-s, 1-s and 2-s.
We now prove that this is not the case when n=3k+1.
Proof 1. We note that the labelling is preserved by rotation around the centre with angle 120∘. Hence by splitting the honeycomb minus the centre into three parallelograms of sides n=3k+1 and n+1=3k+2 as above, we note that the numbers of 0-s, 1-s and 2-s in each parallelogram must be one third of the total numbers of 0-s, 1-s and 2-s in the honeycomb minus the centre. However, the number of hexagon centres in such a parallelogram is (3k+1)(3k+2) which is not a multiple of 3 and so cannot contain the same number of 0-s, 1-s and 2-s. Hence the numbers of 0-s, 1-s and 2-s in the honeycomb minus the centre cannot be equal to each other.
Proof 2. As we have seen above, the region obtained by removing an n-level honeycomb from an (n+3)-level honeycomb can be covered by trexes without overlap. This implies that the numbers of 0-s, 1-s and 2-s in an (n+3)-level honeycomb (with centre removed) coincide if and only if they do so in an n-level honeycomb (with centre removed).
Because in case n=1 there are three hexagons labelled 0 but no hexagon labelled 1, the numbers of 0-s and 1-s in a (3k+1)-level honeycomb (with centre removed) do not coincide.
Proof 3. For each a∈{0,1,2} and each n, let hn(a) denote the number of hexagons labelled a in Hn, the n-level honeycomb (including the central hexagon). We wish to find hn(a)−h0(a). Let
cn(a)=hn(a)−hn−1(a)=bn(a)+vn(a)
denote the number of labels a on the collar Hn−Hn−1, where vn(a) counts the labels at the 6 corners and bn(a) counts the remaining vertices.

We can compare Hn−Hn−1 with Hn−2−Hn−3, as the labels at the outer border (except at the corner) repeat the labels from the inner border. Hence the following recurrences for n≥3:
bn(a)=bn−2(a)+2vn−2(a)and so
cn(a)=bn(a)+vn(a)=cn−2(a)+vn−2(a)+vn(a)
while v3(1)=6 and for n≥4 we have vn(a)=vn−3(a). Also using hn(a)=hn−1(a)+cn(a) we can fill in the following table.
Applying the recurrence relation
3 times for
n≥7:
cn(a)=cn−2(a)+vn(a)+vn−2(a)=cn−4(a)+vn(a)+2vn−2(a)+vn−4(a)=cn−6(a)+vn(a)+2vn−2(a)+2vn−4(a)+vn−6=cn−6(a)+12as
vn−6(a)=vn(a) and
vn−4(a)=vn−1(a) and
vn(a)+vn−1(a)+vn−2(a)=6for all
a∈{0,1,2}. Hence by induction we can prove
hn(0)hn(0)=hn(1)−1=hn(2)=hn(1)+2=hn(2)for all n=3k and 3k+2,for all n=3k+1.Hence after removing the central hexagon, we have shown that the n-level honeycomb cannot be covered by trexes if n=3k+1, since at least 3 hexagons labelled 1 would remain uncovered.
Final answer:
All n except those congruent to 1 modulo 3 (i.e., all n such that n≡1(mod3)) can be covered by trexes as described.