One thing I find incredibly useful is the MathSciNet reviews. Often the reviews make the paper more clear. If the paper doesn't address the big picture, the review will often give context or, if the result fits into a larger program of research it will mention this. Sometimes the review highlights unique technical arguments.
Also, everything is hyperlinked, in that you can pull up all articles referenced by a paper or, better, all articles (or reviews) that reference a particular paper (this has often been useful when I didn't understand something in a paper or didn't know the point of a paper). I spent a lot of time reading reviews as a beginning grad student and it helped me get a sense of what questions people find interesting.
