Should we accept a symmetric accessibility relation for possible worlds semantics within metaphysics?

Possible worlds semantics is a common tool for doing metaphysics. An interesting question is the question of which properties should be ascribed to the accessibility relation between the possible worlds. Now, surely, we need reflexivity: each possible world can access itself. On standard Kripkean semantics, this guarantees that necessarily true propositions in each world are also true in that world, which is a principle without which possible worlds metaphysics becomes senseless. Moreover, we need the accessibility relation to be transitive. After all, if possible world w2 is possible from the perspective of possible world w1, that is, if w2 is accessible from w1, and if possible world w3 is possible from the perspective of possible world w2, that is, if w3 is accessible from w2, then, plausibly and reasonably, possible world w3 is also possible from the perspective of possible world w1, which means that w3 is also accessible from w1. Indeed, transitivity guarantees that necessity is stable or robust: if a proposition p is necessarily true in some possible world, then p is also necessarily necessarily true in that world. Now, what about symmetry? Should we also require that the accessibility relation be symmetric, meaning that if possible world w2 is accessible from w1, then w1 is also accessible from w2? This doesn't seem to be the case. Here is a proposed counterexample. Consider a possible world w1 and consider the conjunction p of natural laws in w1. Suppose that in w1 the natural laws are necessarily true, so that p is necessarily true in w1. Thus, p is true in each possible world accessible from w1, which includes w1 itself due to reflexivity. Now, consider a possible world w2 in which p is false and whose conjunction of natural laws, say q, thus differs from p, and in which the natural laws are merely contingently true. So, there is at least one possible world accessible from w2 in which q is false and whose conjunction of natural laws thus differs from q. In this example, since p is false in w2, w2 is, given Kripkean semantics, not accessible from w1. But plausibly, w1 is still accessible from w2. For while the natural laws in w2 are contingently true and have q as conjunction, it's from the perspective of w2 reasonably still metaphysically possible that they could have been necessarily true and have p as conjunction. That is, w1 is possible from the perspective of w2. In this specific plausible case symmetry thus does not hold. It follows that we should not accept that the accessibility relation is symmetric. Therefore, the most adequate modal logic for doing possible worlds semantics within metaphysics is S4 (reflexivity and transitivity) rather than the stronger S5 (reflexivity, transitivity and symmetry) or the weaker KT (only reflexivity).