(I think the App Store gets a lot of ink because people have this notion that Apple does the marketing for devs like you. Which is true, in approximately the same sense that the MBA does marketing for basketball players like Shaq.)
For SaaS, it's also important to model things like trial conversions, turnover, etc. which don't appear in a single-sale case. There are a lot of differences. But the odds still seem better there.
And of course I am completely overlooking the "what about ads" question, mainly because you need a ton of installs before that starts being a significant amount of money. Better than not, helps feed conversions, but doesn't change the equation for me.
That said, the students loved it. And, he could also update the app as the course went along. Finally, although he only made enough through App Store sales to maybe go out for a nice meal, it translated into promotions, bonuses and the like. So ultimately, it paid off.
My main point is: in addition to selling niche apps at a premium, there's also the opportunity to develop cheap/free apps that otherwise help you move ahead within a niche market.
To succeed, it seems that you now either need a mountain of luck, or a killer marketing engine and a fair amount of luck. Mathematically, it seems unattractive compared to other ways of applying the same time and skills unless you have something you're obsessed with building and want to produce irrespective of the outcome probabilities.
E.g. a $1.99 app pays at $1.393, so you need to move 34893.75 paid copies in order to clear a $50,000 gross. For SaaS and the same target, you only need 208.33 customers at $20/mo. The capture difficulty is not identical between cases - but if you have good marketing and are assuming serendipity on either side, it's quite feasible to hit 208 customers. Furthermore, it's often easier to grow once you get there, and it's quite possible to charge at much more than $20/mo/customer.
Mobile apps are great, but not a "for profit" activity for me. I'm all for doing them for fun or ad hoc, and if that takes off - great. But it seems to be a very difficult business model compared to applying similar resources elsewhere.