This is one reason why GPA carries less weight; there are just too many variables. Quality of school, difficulty in gaining admission, competition, and grade inflation differ for every school and every major. Then they must consider how long someone has been out of school, whether they worked during school, perhaps if they changed majors. They account for GPA within a major, GPA trends and even if someone takes classes after graduation.
It's very inexact. Two people with 3.3 GPAs could be vastly different candidates. The first could be at a school that is extremely difficult to get into, say Berkeley, where competition is extreme. Maybe he spent two years in pre-med where his GPA was hammered before changing to be a engineering major or something. Perhaps he worked 30 hours a week during the entire time. You'd have to consider that first GPA way differently than another person who's at a small local school where entry is non-competitive, grading is lax and their way was paid for by their parents.
I had a friend at Pepperdine. I asked her about a mid-term she had and she said she did poorly because she didn't study. It was alright though, because the entire class did really poorly and the professor decided to drop the grade. First of all, I'm sure this would have never happened at UCLA (where I attended), or any top competitive school; but if it did in fact happen, they wouldn't have a problem just flunking everyone. The level of work and the competitive environments are simply not comparable.
So, I believe that GPA plays a relatively small roll in the overall process, and they give you plenty of chances to offset any problems. You can dissect your GPA and present an upward trend, or good grades within your major, or good grades in quantitative classes, take courses online or at a community college, or even point out that you graduated a long time ago and are a different person now. Too many variables means very little certainty, which limits the usefulness as a data point.