Sounds easy to evade. Once a machine has recorded 10+ votes since the last time it outputted a count, it could start giving a small probabilistic advantage to a preferred candidate, at the expense of other candidates. There would be no bumps, and the results would never differ by more than, say, relative 20% from the real count. E.g. if preferred candidate got 10% on one voting machine, it would give him 12%, taking the other 2% from all other candidates.
No bumps or disproportionate popularity in traditionally unpopular regions to raise suspicion.
No bumps or disproportionate popularity in traditionally unpopular regions to raise suspicion.