I'll try to adress several issues and misconceptions that I noted in the above posts.
1) Averaging the measurements as you are doing is an excellent method of reducing the impact of the within part variation (ovality). Variation is reduced by the square root of the sample size (SQRT of 3 = ~1.7), so you have reduced within part variation by almost half.
2) Your operators are very close in their measurements. The Operator portion of Reproducibility is 0. The issue is the interaction between part and operator that was caused by part number 2. Have both operators remeasure part number 2 and reanalyze the study with the remeasured part data.
3) justncredible, There were 2 operators. The 1 that you saw was the degrees of freedom for two operators (df = n - 1 = 2 - 1 = 1)
4) The number of distinct categories is correlated to the %GRR, but does not cause it. The two are different metrics measuring the same thing.
5) ANOVA is a better method than the Range method used in most spreadsheets. It provides the operator x part interaction, which the Range method does not provide. In addition, the Range method uses the range to estimate the variation versus the direct calculation of variation by ANOVA. Stick with Minitab over the spreadsheets.
6) Regarding the ndc and %GRR: These metrics are highly dependent on the variation of the 10 parts used in the study. They must accurately reflect the actual process variation. 6 standard deviations of the parts used in this study were 0.0003576. How does this compare to your process capability measured the same way (average of three)? If it is smaller, your ndc/%GRR will be worse than they actually are.