How much money is NOT being made by NOT publishing stuff that's still under copyright but that isn't profitable enough to pay royalties?? (And maybe isn't profitable enough to justify tracking down an absentee copyright holder.)
Clearly there IS money in publishing old stuff, or most of the pre-1900 classics would be long since out of print, and such is not the case. They continue to be reprinted to this day.
I would guess that over the long haul, long copyrights result in a net reduction of money to be made all along the chain -- remember it's NOT just the author and his agent and the first publishing rights, but also all the reprint houses, distributors, and bookstores. It occurs to me to wonder how much long copyright contributed to the demise of small local bookstores, and may now be contributing to libraries that are social hubs but no longer house vast numbers of books.
I think that a long copyright hinders the creation of new material simply because it may be too close to something already under copyright. For example, a writer is considering writing a story featuring a spy, but decides not to because the spy/story would be too close to an already existing spy/story that is in copyright.
A somewhat recent example of how copyright can hinder story creation is in the series "Star Trek - The Next Generation." In an early season they did an episode featuring Dr. Moriarty from the Sherlock Holmes stories. Due to copyright issues they had to wait years before they could do a follow-up story featuring the character.
I think a consistent and predictable influx of published material into the public domain will, in the long run, lead to the creation of new work. I do think that the rule should be simple and consistent, rather than having several rules (such as extensions) which lead to a situation of "this story which was published in 1950 is in copyright, while this story, which was also published in 1950, is in the public domain."
On the subject of how long copyright should be, rather than having copyright rules which allow for variations (as mentioned above), I think that a better way would be to have a copyright of 30 years (basically one generation) from the date of publication, rounded up to the 1st day of the next year (for example, by this rule anything published in 1979 would fall into the public domain on the 1st of January 2010). This would make it easy to determine what is and is not in the public domain, and would ensure that works do fall into the public domain in time.