Study 1 is across college-aged kids who are willing to participate in a psych study (in exchange for partial course credit or a lottery entry for a $40 gift card, IE they're psych students). Unclear why you'd generalize a study run across a single college, within a group of students who self-selected into your course, and draw conclusions about all men and women. In addition, you're asking people what they prefer in their friends; not measuring it.
Study 2 isn't controlling for variance in the underlying traits between the genders. All it shows is that if your best friend is male, they're likely to have different traits than if they're female. It does not show that you picked male/female friends because of those traits. So for example, when they find "men's same-sex best friends were more likely to possess qualities of physical strength", what they've discovered is not that men look for physically strong friends, but that men are usually stronger than women.
Study 3 is across people working on Mechanical Turk. That already skews your sample. It asks participants to weight the relative aspects of what they look for in a friend. But this relies on the participant being aware of what they look for. If someone thinks that they don't need emotional comfort from friends, they'll say so, but it doesn't mean that it's true.
tl;dr: This study is methodologically flawed, and the conclusions it draws are mostly to be splashy and show up in random articles. It'll be shared because people resonate with its conclusion and not because it contains robust evidence of its conclusion.