This is a very good point, the curve-scaling of the votes the way it is now that alters the payout time allows the big votes that come in to progressively extend the payout while other big votes (or downvotes) could still come. The linear distribution does mean that you can gain more on a big one than a little one so there is good reason for the time scaling.
Having said that, with the proposed shift to 7 day payout cycles you don't have this issue, and the votes that go to other than the ones already with big rewards slated are slowly diminished if other posts rise up past it, which counters the avalanche tendency.
But like someone above said, it would be really useful to take historical data and model the distribution of a different algorithm to actually see how it would have affected the payouts already past. Hopefully someone will build that soon. It would also be helpful for the idea that I am promoting, that of witnesses being able to set charges instead of dividends on SBD holders.