I know - the whitepaper is very weird and misleading on that point. I had to reverse-engineer the algorithm myself from the code before I actually believed this, since it's not documented anywhere. Each time you vote, that gallon of gas you take out of the tank all goes completely to upvoting the post; none of it gets wasted on something else. Your car is just as efficient with an empty tank as with a full tank, it's just slower (might as well beat this analogy to death). It's exactly this:
power_after_vote = power_before_vote - used_power
where used_power gets multiplied by slider power and SP to calculate the rshares that go to the post.
That exact expression shows up in this line of code.
OK, i see what that graph is saying now. Wow... ive actually bee spacing them all this time.
Incidentally, based on this, I made a model... tell me if you think I'm on point in my analysis.
https://steemit.com/steem-help/@sigmajin/the-tale-of-the-5-brothers-a-voting-power-parable