It would be interesting to do a performance shootout of all of these, but for my money @SteveXD’s original solution is the most elegant to read, if maybe a little opaque if you haven’t internalized what spread and gather do.
I was thinking about the conversation in this thread which touched on the idea that it’s better to take advantage of R’s column-oriented-ness than to swim upstream with rowwise operations. I’m not completely sure how well this situation connects to those ideas, but it strikes me that maybe converting the problem to a colwise operation with a minimum of steps is a major point in favor of @SteveXD’s original solution?