I never really know how much time to spend on take-home problems. I could spend a lot of time on it and it should be better -- possibly demonstrate more of my skills -- but often that ends in it being over-engineered and ultimately bad or otherwise unrepresentative.

Getting paid could provide the ideal constraints for the problem at hand.