The discussion over whether in-socket thermistors are precise and reliable or not has been questioned many times. I have put 2 different motherboards with 2 different CPUs with their in-socket thermistors up against an outside thermal probe from a Compunurse. A guide can be seen here on how to install a Compunurse to a CPU. The Compunurse in both test setups is directly touching the side of the core as which can be seen in this picture.
Test Bed #1
Artic Silver III first let the computer idle for about 20 minutes to get a baseline reading. As you can see the Compunurse, which is directly touching the side of the core, is 3.5 degrees higher than the in-socket thermistor.
I then used Prime95 and torture tested it to get the CPU at load. Again the socket reading is 4.6 degrees below that of the Compunurse.
Test Bed #2
As you can see from the above tests and results you can see that the in-socket thermistor and Compunurse give quite different results, also the 2 different motherboards give off opposite results. I believe that in both cases the Compunurse is the more accurate reading because of the fact that it is touching the side of the core on both CPUs while the in-socket probe is under the CPU and not touching it whatsoever. Because the results from the two motherboards VS Compunurses are opposite I believe that the MSI motherboard has a more accurate probe compared to the EPoX. While they are both fairly close I would always learn more toward the side of the Compunurse.