Natural Language Processing (Dan Jurafsky, Chris Manning) was the one that took up the most time. I felt like to get a perfect score (which I got nowhere near close to) would have easily been 10+ hours per assignment (not to mention actually watching the lectures / quizzes). This may have been in part due to my own lack of background knowledge (up until coursera I was almost entirely self taught). But even Algorithms, Part I (Robert Sedgewick, Kevin Wayne) felt like it took at least twice as long to get through the quizzes + assignments as Odersky's class.
I took both and I confirm Scala course was less work than NLP. However that might be just because NLP course had much bigger and more tweakable tasks - on some algorithms, you could spend literally days to squeeze one last percent of accuracy out of it (and then end up overfitting, failing the hidden tests and having to start over again). NLP tasks weren't actually general programming tasks but more NLP-specific work - which is great, but it doesn't compare to generic Scala course. Scala tasks were much smaller and well-defined. I probably could use a bit more complex tasks, but being complete Scala novice before the course, I couldn't really - at least yet - do some heavy lifting with Scala as I could, say, with more familiar to me languages like Python or, say, C.
As a fairly experienced programmer, I spent about the same amount of time on assignments for Algorithms, Part 1 and Odersky's Scala class. I think the main difference was the lack of standalone quizzes in the Scala class, but at the same time I'm not sure that having quizzes would've added much in terms of additional understanding.
