depends what the actual distribution of the whole numbers (not just leading digit) looks like. The "Benford's Law" distribution for leading digits would be expected for distributions with uniformly distributed logarithm over some orders of magnitude. For more narrow distributions, it shouldn't hold.

In the event that these numbers are, say, the number of voters per geographic zone, and these zones are similar in number of total voters, and the voting habits in each zone don't vary a lot, it makes sense for the less popular candidates to more closely follow Benford's law, because the number of votes is smaller and proportionally more random. Errors add in quadrature etc.

Does anyone have the data that was used to produce this? I'd be interested to take a look!