Looks like Minitab uses the bias-corrected estimator with the "-3" convention applied --
K>> kurtosis(x(:,2),0)
ans =
0.63
K>> 3-2.37
ans =
0.63
K>> 0.63-3
ans =
-2.37
K>>
With the default for the optional flag default in MATLAB (==1), then you get the result you got.
See the 'Algorithms' section of the documentation for full details.
With only 5 observations and that the estimator of K uses sum((x-xBar).^4), the random variation in the estimator is quite high and I wouldn't put much faith at all i such a statistic as far as making any judgement from it.
