Alot of theories, mostly the less rigorous ones, rely on group selection. But the strongest ones rely on classic genetic selection where homosexuality does directly benefit their genes, and specifically the homosexuality gene. For example, one well-known experimental study looked at siblings and, IIRC, found that the sisters of male homosexuals were more fecund. One of the theories was that a gene which in men promoted homosexuality had the effect in women of promoting reproduction by increasing sexual attraction to men. Genetic selection through survival and reproduction principally acts on genes singularly, not the whole animal, let alone species, which are derivative effects that we too often conflate with the core dynamic. Of course, an alternative explanation in this case might be that male homosexuals can help provide more resources to their siblings, which given the degree of genetic relatedness doesn't require relying on a group selection effect; but it's more tenuous and less plausible than the explanation relying on a very straight-forward selection effect directly increasing replication success of a specific gene. Reproductive success isn't about whether the specific molecular copy of a gene replicates successfully through a lineal chain, but the success of any copy of itself, anywhere, no matter how distant from a shared meiosis event (or, in principle, any shared meiosis event).

The wandering penguin notion can be analyzed in the same way. A gene that induces wandering, which in all but an utterly minuscule number of cases results in a dead-end, may superficially seem to be counter productive when judged in isolation. But is it? How are new colonies formed, and who (or what) benefits if and when a new colony grows and thrives? Not just the species, but the specific wandering gene will see massive reproductive success as the new colony grows, at least initially.

Of course, a "gene" is an amorphous thing, and intuitively wandering needs to be attenuated, so maybe the relevant "gene" here isn't just something that makes them wander, but the whole package of DNA that also encompasses regulation of propensity as manifest through the population. But we don't necessarily have to cheat that way, either. Most of the time the wandering gene would be a net negative and find itself slowly winnowed out of the population. But as long as it survives somewhere in the population long enough to induce new colony formation and benefit from a short explosion in reproductive success, it'll survive, at least globally. Heck, maybe in long-established colonies it disappears completely, only to be reintroduced by wandering penguins from younger colonies. We don't need to zoom out and model how it interacts with all the other genes until we've zoomed out so much we end up in the position of positing a group selection effect. That's the beauty of Darwinian genetic selection--all this complexity arises from a very simple dynamic that in almost all cases can be accurately and predictively modeled by just looking at specific genes in isolation, fundamentally independent of the species and, strictly speaking, even of kin groups and individual animals; and what exceptions do exist don't require zooming out nearly as much as people tend to do.