The tricky part is setting up your pipeline: from .csv files to submission. This can take a day, or 30 minutes, depending on the contest. If feature engineering is required then this adds a lot of time to this process.
Time to create the first submission dropped drastically after a few competitions. I now have a small library of munging and ensembling scripts that I can quickly adapt to suit the needs. On the other hand, time spend on optimizing and staying inside top 10, increased too. For the KDD-cup I'ddo weekend long sprints for a few measly improvements. All in all I'd say I spend at least 8 hours per competition.
My background was front-end developer growing into analytics and dataviz more and more. I think it was on HN that I saw a link to learnpythonthehardway.org and I started from there. After reading "programming collective intelligence" I got more serious.