I noticed the reason the results look weird is because it is measuring the bar based on percentage rather than the number of responses. I propose you make it so that the bars measures the number of responses, rather than the percentage.

This would mean the selection with the most responses would span across the entire thing, and the other selections would be compared to it.

The percentage method does not make good use of space and is a really bad comparison.

Or perhaps we could have the option of using more than one kind of graph?

Pie Graph for Pie Day!

Oooh, what about a line graph that has the number of posts on the X axis and (likely numerical) options on the Y axis, and people have the option to change their vote at any time, but the new vote would be recorded at the current post number? Like, in the event of a debate thread where people get their minds changed after really good posts? Or am I thinking too much?

Hmm... good ideas... but I don't think they'll change the graph type.

Everyone's ideas are great. The way we can combine them all, is with choice: you can choose how you want your graph to look like, and how you want to graph your poll results.