With my smaller subset, no problem. Perhaps, a Data Loader dialog could be used here like what you have done for the Semivariogram/Covariance Cloud. Instead of saying that less is more to find the interesting data, it could say, if you use more than 1000, it may crash.