This post comes from notes I jotted down before I spoke, so it may not reflect actual content, whereas I think I basically captured what other people said. (It's come to my attention that some people consider my notes on other people's presentations freakishly complete. I blame eight years of policy debate.)
Most of my experience with orphan works comes from contacts with historians. Images are the biggest problem: you may know where you found them and sometimes who's in them, but rarely who created them. We have a family photo from early in the 20th century that shows members of my husband's family posing with their favorite socialist newspaper. Another historian wanted to use the photo in a book about Jewish politics in New York; we sent him a copy and gave permission. Did we have any right to give permission? Who knows? The use wasn't conventionally transformative but may have been a fair use – the historian was using it to illustrate his argument about the importance of small differences in politics by showing that the choice of newspaper was significant enough for a family to use it as a means of self-definition. Regardless, the publisher wanted the empty ritual of permission.
More generally, authors are hard to trace, and if anything the problem is worse when the copyright owner was an organization that's now defunct. The successors often don't know what they own or if they own it. They worry about giving permission, so refuse or don't respond to inquiries, or they blithely give permission without owning the rights.
A story from the heirs' side: My grandfather, Leonard Tushnet, wrote short stories and a few books. They were all published after 1963, so renewal was automatic; he passed away before the first terms ended, so the renewal terms went to his heirs (my father and aunts). To my surprise, I found one of his stories reprinted in a recent anthology. I'm sure the publisher sought permission from the original publisher – and got it – in good faith, but it was nonetheless mistaken, and strictly liable. The heirs have no plans to enforce the copyright, but I tell this story to point out how easy it is even for repeat players to screw up.
The basic question I'm interested in: How hard should we make new users work? Carnegie Mellon invested several million dollars in digitization, a substantial chunk for permissions. Very roughly – read more here or here -- they found 1/3 of the publishers fairly easily, 1/3 with difficulty, and 1/3 not at all. (And, as noted above, some of the ones they found didn't necessarily have the rights.) Is this game worth the candle?