Dear Juergen! This reminds me of my very old tests when I run Spyder 3 for 8 hours reading my monitor white. Caution! This data is very old (my original post is dated on Sat, 7 Mar 2009 13:41:37 +0100). I prepared a chart similar to yours - the Median of [XYZ] is a median of the whole dataset (8821 results), not only the first 2000 shown on the chart (the rest 6000 are about equal to 2000th). It seems that if you had waited a bit longer the results would have flattened out! :). The possible explaination to this phenomenon is the thermal stabilisation of the Spyder. The monitor has been warming for about 8-10 hours before the test (I had been working with it and run the tests for the night). Lukasz.
Description: GIF image