1. If you ran prime 95 for 8 hours, stopped it for a min, and then started it again for 12 hours. Is it considered 20 hours stable or 12? What if you stopped it so you can reboot and drop the offset a notch? I suppose in this case it's a bit different since you're working with new voltages.
That is not a 20hr run, that is an 8hr test followed up by a 12hr test.
The reason the test times are not additive is because of the underlying nature of instabilities in the first place. They are non-linear.
For example, an unstable system will not always fail at the exact same time during a looping stress test. An unstable rig might pass an 8hr stress test only to crash after 8hrs and 1 minutes. Then tomorrow you run the same stress test and it might crash after 2 hrs and 10 mins. Another day later it might make it all the way to 15hrs before crashing.
Repeatedly passing a stress test of a given duration (8hrs or 12hrs or 24hrs) is also no guarantee that your rig is stable, it just means that statistically you can be confident that your mean-time-to-instability is much larger than the maximum duration tested during your stress testing regiment.
Passing a stress test once, for a given specific duration (i.e. "my rig passed IBT for 12hrs, must be stable") is not valid outside the enthusiast community.
Stability is a matter of statistics and sampling size, not singular corner case checkpoints.
2. If during stability testing, some applications stop working but the PC doesn't crash/BSOD, is it considered unstable? I get these popups during IBT with windows explorer and antivirus services failing, etc. I assume it's because the PC is under too much load to allocate resources to it?
You should be so lucky as to have such an obvious and unmistakable tale-tale sign of system instability as having an application actually stop functioning.
The more worrisome instabilities are the kind that cause silent data corruption, your files on your hard drive slowly become trashed (windows stops loading correctly after a while, necessitating a clean install, your personal files become corrupted, zip files won't unzip for CRC errors, etc).
Yes, if something doesn't work right then it is not working right.
3. Do you use your PC during testing? Is it OK or does it mess with testing. I assume it's ok since you're putting even more load into it. I don't do it with IBT though since it does seem to be under complete load[PC is sluggish] but with Prime 95, I was able to surf the web with Firefox at the same time without any kind of lag.
Stress testing requires the undivided attention of the resources being stressed. When you run parallel applications to your stress tester those applications are consuming cache and ram, not to mention CPU cycles, to the detriment of the efficiency of the stress testing program.
Leave your computer alone while stress testing, if you don't then you risk reducing the intensity of the workload placed on the CPU (or ram, or GPU, whatever component you are testing at the time) by the stress tester in question and in doing so you risk generating a false positive that your rig is stable when it actually might not be stable.