...What? How? Why? Who? How do these people keep getting work?
Twice in a day I find myself in a bewilderment.
40,000 votes later:
At the suggestion of animateclay, I looked at the HTML of the graph. There's a max-width:300px CSS declaration on the .ans_bargraph elements (the green bars) that's overriding the accurately sized percentage-based width declarations: Ron Paul's bar was at 100%, and Romney's was at 49% width when I checked, but RP's 100%-wide bar was being capped at 300px...which just so happened to look exactly like 50% in a 600px wide chart.
So, in short, at best, (ignorance before malice, of course!) no one tested their chart code ahead of time with any good data sets. Disappointing because the consequence is so deceptive.