It could also have surfaced in a different media outlet, prompting many to see it for the first time. For example, some article comes out and gets submitted, but the submission dies out. Then a blogger or news site mentions is, and people re-submit the article to HN all in a short time (but much later than the original submission)- which counts as an upvote.
Also, the first abnormal result the OP was talking about 3 votes. I don't think we can worry about a handful of people voting for their own startup submission. The first, "normal" graph had hundreds of votes.
I think my takeaway from all of this is that, like other statistics, with small numbers the values have little meaning. Now if we had seen evidence of, say, 25 or 50 people working together (a few people with multiple accounts each, for instance), then that would be more actionable evidence. But my guess is that pg has already dealt with that problem, and apparently pretty effectively, because it's not showing up here.