We use a simple kurtosis formula in Geostatistical Analyst. The formula is the fourth moment divided by the square of the second moment. We do not correct for small samples or calculate excess kurtosis.
It appears that the KURT function in Excel calculates excess kurtosis and corrects for small sample sizes. In general, the difference between these two calculations should be approximately 3. You can see from your data that if you add 3 to the Excel values, you will get close to the ArcGIS values (the remaining difference is due to small sample corrections in Excel).