This is my interpretation: The distribution is skewed because these groups actively seek out the top interestingness photos. It's not a case of being in those groups that will make your photo more interesting. In other words there's no causal relationship. Does that make sense?

Yea, I hear ya. We can't distinguish between the groups the photo was in before it hit Explore and the groups it was added to after it gained attention for its Interestingness.

"Does that make sense?"

yup!

although certain groups meant to collect the top 500 explore photos have a very low percentage.

this could just mean that they wait more than a day to start collecting, but it it still interesting to note.

It's hard to tell based on how he describes it, but it sounds to me like he's looking at the most recent days' top 50 interesting photos. So, of course, any group that requires your photo to be in Explore will not be highly represented simply because the photo only got to Explore that day and the owner hasn't had a chance to add it to those types of groups yet.
I think what would be more interesting is if he took, say, a date 6 months ago and looked at the groups THOSE top 50 photos are in. Then we'd see the full effects of group membership.
Or am I misinterpreting his data collection method?

"Or am I misinterpreting his data collection method?"

i don't think so. he should probably offset collection by a day at least.

What would really be interesting is if he could predict 'most interesting' photos by checking up on which groups the photos are in and what kinds of cameras are producing them.

As I explained to my son yesterday, "You are a boy. You go to school. Therefore everyone who goes to school is a boy."

If knowing the extra info he mines about the photo does not let you predict or give probability of it hitting Explore, then it probably doesn't really mean much by itself. Notice that currently 4% of Utata photos hit the mark but 8% of FickrCentral hit it. My unsubstantiated guess is that winning photos are posted to several of the top groups while Utata has a five group limit.

Not only are the top groups "invite only", all but one also require posting a gif in comments, and two actually require faving photos in their pools. If the interestingness formula takes into account the source of the fave or comment, or the size of the comment, things may be recursive. At some point interestingness feeds on it own output. I've rambled long enough as my comment may be larger than the original article.