I was just reading about that issue in the Intel documentation the other day. They didn't point the finger at any particular program but were sort of hinting at that issue and programs that need to read data from model specific registers on a regular basis. Maybe I took that info too personally.

My opinion is that even if you stopped running RealTemp, there would still be other background activities that would constantly wake cores up and out of the C3/C6 state. RealTemp isn't a huge load on the CPU but it does need to access each core once every second to query the temperature data for each core.

I'm not sure how you could test for this. You could try turning RealTemp off and then wait a minute and use something like my MSR Tool and try to read the sensors directly in MSR 0x19C without having to start RealTemp. Maybe something like a Kill-a-Watt meter might be able to show a slight reduction in power at the wall but I don't think it will be possible to see any measurable difference of temperature or power, with or without RealTemp running.

I thought about adding an option to RealTemp so it would be possible for it to not read the temperature sensors when idle but then I'd have to query the CPU to see when it's idle.
It's a no win situation.

I have a fairly lean system so at idle the RealTemp C0% based load meter gets down to about 0.6% with one or two instances of RealTemp running. Even if RealTemp wasn't running, I think there would still be some background activity running on each core every second.

My thoughts are that if you are running a well overclocked i7 900 series CPU, RealTemp is probably the least of your concerns when it comes to it heating up your core temperature or increasing your power bill.