Any special reason why you would want to use kriging on lidar data?
Kriging will take forever when you have a large number of points and it may never finish. I would suggest you look into less memory intensive interpolators. Kernel Interpolation (with Barriers) in Geostatistical Analyst is an example of one that I have used with roughly 4 billion input points.
I was trying to use kriging because I have a second lidar dataset for the same area (different time) that was interpolated that way, and wanted to use the same method to minimize the uncertainty that it adds to the differencing between the two surfaces. It's only (that sounds ridiculous) half a million points or so, so I wasn't expecting it to take so long.
I'm going to stick to IDW for now. The error that the change in interpolation method seems to be adding is tiny compared to the accuracy of the dataset.
In some circumstances, given the interpoint spacing of lidar data (often 2 m or less) there is no point in doing an interpolation in the first place and simply converting to a grid produces more than suitable accuracy
