In order to share valid data, we need to clarify a couple of terms. I have no personal experience or data with the practice of power-cycling servers on a scheduled basis. I do have personal experience and data with re-booting servers on a periodicbasis.
What I can tell you is that our data shows that most of our lower end servers will have an "out of virtual memory" event within a 30 day period if not re-booted within that period. Keep in mind that these are machines that are severely memory constrained (64meg in some cases) to begin with.
I don't disagree that memory leaks should be amenable to correction, but unless you're willing to hack your OS, you have to wait for the vendor to get around to your particular issue. The three components most often guilty of creating memory leaks on the Compaq/NT platform have been SNMP, the Compaq utilities, and anti-virus software and at the applications level, we don't have control of any of those components. A periodic re-boot is the workaround until a fix is available.
Our data does not suggest that periodic re-boots have a deleterious effect on hardware reliability but I would be hesitant to say it would apply to power-cycling as well.
Also, just as an aside, I have Netware 4 and Unix servers that have been up for over a year (in one case, over three years) with no re-boots. And my initial experience with Windows2000 is that it is dramatically more stable than NT4. Ironically, my experience with SP6a has been farworse than with SP5, our current standard.
So, for the time being, we do periodic scheduled re-boots of our NT servers during off hours to minimize the need to do so as a surprise event during peak hours.
paul