I don't see any reason to believe that the total vote, V, is a fixed number divided into two parts, let us say R and L, such that R + L = V, and that then the L portion, also a constant or at least a mighty sluggish variable is then divided into two other parts, let us say G and N, such that for every increment to N there is a decrement to G and vice versa. This view is simplistic and mechanistic.
In 1992, Perot was supposed to have somehow loosened up votes from Bush which then floated over to Clinton. I think this sort of dynamism is closer to the phenomena we actually observe in polls and anecdotally. That is, the appearance of Nader may cause votes to move not only to him from either Gore or Bush but from Bush to Gore or vice versa. People may decide to vote who were going to stay home, or vice versa -- Nader does give the election, otherwise crushingly boring, a bit of interest. Other voters may become interested in yet other marginal candidates because Nader's candidacy gives them ideas.
Another good example might be 1948. Truman should have lost, because both the right and the left pulled out of the Democratic party and ran competing candidates. In fact the situation must have made Truman more credible to the voters.