In this paper, the main attention is paid to the analysis of Windows operating system counter Memory Cache Bytes. Thanks to the Windows Perfmon tool, it was possible to gather long time series that show interesting, from statistical point of view, behavior of different Windows Desktop operating system versions (Windows 7, 8, 10) in idle and loaded mode. The comparative analysis of this counter behavior, understood as a time series that represents cache memory, will show that based on internal memory management mechanisms power law distributions are omnipresent phenomena. We focused on McCulloch, Kout methods and Hill estimator for calculations of alpha stability index and Q-Q plot with Anderson-Darling test for distribution normality test. All of the tested time series indicated the existence of deep probability heavy-tailed distributions for extreme values, confirming that operating system has to deal with anomalous cache memory behavior. This feature is common for all tested 64-bit hardware configurations regardless of the workload mode.