There are three answers to this. I'll spot you the first two:
1. "As long as it needs to be to compellingly and entertainingly tell the story, and no longer."
Can't really argue with that--given the story is worth telling, and stipulating the compelling and entertaining part. If that takes two hours, then it takes two hours. BUT...
2. Just as 80% of us think we are above average, most of us think we are more entertaining than we really are. So while the populist answer is "as long as it needs to be," the practical answer is probably "shorter than you think." Proponents of the "shorter is better" theory often cite Stitcher's 2014 data that the average podcast listening session (on their platform) is 22 minutes. That stat, however, is likely extraordinarily misleading. I'd rather see the median session time than the average--while the average might be 22 minutes, I really doubt most sessions are 22 minutes. I'm betting most sessions exceed 22 minutes, with a smaller number of really short sessions pulling the average down. That's a guess, of course, but an educated one. :)
I'll give answer #3 in a bit (it's my current answer) but I will say that I have historically leaned towards the shorter-is-better camp for the most part, though my favorite podcasts do in fact tend to be longer--but that's inductive reasoning from a sample of one. If your Westworld podcast is longer than an actual Westworld episode, I dunno. You'd better be Anthony Hopkins-good.
Still, a significant percentage of the most popular podcasts are longer than 30 minutes--and many are even longer than 60 minutes. Recently, Rob Walch disclosed some statistics from podcasts hosted by Libsyn under the headline "Surprising Data about Optimal Podcast Length." Walch revealed that over 80% of their most popular podcasts (podcasts with 100,000+ downloads over a roughly two-month period, a fair qualifier) exceed 51 minutes in length. My first instinct upon seeing this data was to posit that many of the top podcasts are either celebrity driven (and fans have a higher tolerance) or are created by public media (who have decades of experience in long-form, spoken word audio and are just good at it.) Both are true, but there are plenty of podcasts in the top 100 that fit neither criterion--they are just good shows.
Now, hopefully no one who reads this blog will fall into a lazy causality trap: Podcast A is Good, Podcast A is longer than 60 minutes, therefore Good Podcasts are longer than 60 minutes. But clearly, with these top podcasts, length isn't isn't exactly hurting them. So, people who insist that a podcast should be short don't exactly have a lot of supporting data to back that up. But there is a real danger in misinterpreting the Libsyn data to be prescriptive. Take a quick glance at these three graphs from this year's Infinite Dial study (really, just a quick glance:)
At even a cursory glance, you will note two things: All three are growing, and Podcasting isn't growing with the same velocity as the other two. Now, I have never believed that Podcasting was going to exhibit the same kind of growth curve as online music--it's a very lean-forward medium, and doesn't lend itself well to media multitasking (also, we consume far more music content than we do spoken word content.) But it is fair to look at the podcast graph in the context of the smartphone and online radio graphs and make this statement: there are a lot of people who could listen to podcasts, who are not (yet) listening to podcasts.
All of this is to say that it may or may not be true that the people currently listening to podcasts prefer longer content, or that longer content tends to do better with today's podcast audience. But today's podcast audience is not necessarily tomorrow's podcast audience. And what got us here might not get us there. Which is why I would reframe the Libsyn data not as the "optimal" length for podcasts, but simply as a summary stat for what the current state of the industry is. I will continue to enjoy longer podcasts. But I also see developments like NPR's upcoming debut of its new, 10-minute news podcast Up First not as flouting the "rules" of podcasting, but as an exciting potential gateway to board new listeners to the podcast train. (N.B., NPR is a client.) Up First does not even have an episode yet--just a trailer--and as of today it was #2 on the iTunes Top Podcasts list. In other words, just the idea of a 10-minute news podcast is currently more popular than This American Life.
What does this prove? Well, nothing, really. And that's kind of the point. I'm not sure any data on podcasting's existing audience suggests where the medium could go. With 61% of Americans 12+ listening to online radio monthly, and 24% listening to podcasts monthly, that's at least 37% of Americans (over 100 million!) who could be regular podcast consumers, but aren't yet podcast consumers. And their version of "too long" might differ dramatically from our version of too long. Which leads me, in a way, to question answer #1, above ("as long as it needs to be to compellingly and entertainingly tell the story") because that answer doesn't include the listener, merely the wishes of the content producer. There is an audience, somewhere, for content of all lengths. Optimizing for the audience we have is a sure-fire way to guarantee that's the only audience we'll ever get.
Which brings me to answer #3 to the question I led off with: how long should a podcast be? Here is the right answer: I don't know. I don't think anyone does.