This is an old question now but as I've been looking into this stuff tonight ...
You are calculating the confidence interval for a binomial distribution and using the lower bound as your "score" of helpfulness. A confidence interval defines a plausible range of values for the true mean given your observed values and sample size.
When the sample size is small (like just 5 people) this range will generally be larger as there is less confidence that the true mean has been observed. As such your lower bound could well be lower than the lower bound for another sample with a lower observed mean (like the first item in your example).
So yes, I think it probably is correct (although I haven't checked the answer myself!)
