OK... I've got your exact chip and ran into exact the same problem at 3.8
It took me a while to figure it out. Now, every piece of silicon is different. So my solution might be different from yours... So i'll direct you first to where i learned enough about my chip to overcome the problem on my own.
http://www.overclockers.com/forums/showthread.php?t=596023
Before i tell you how i overcame this wall, understand... that while i was able to get over 3.8 temps and power became a huge issue... and though i was able to get stable overclocks up to 4.0 (unstable up to 4.2), the temps got so bad i ended up pulling the whole overclock back under 3.8 to stabilize temps and give me a rock solid overclock.
So while you might overcome this, you might not get temps your comfortable with under an extended prime95 burn.
Ok... onto the solution.
At about 3.8 you need to bump the HT/CPU-NB speeds, the cpu-nb should be up to a min of 2400, i found i couldn't stabilize the overclock without this, i also needed to bump the voltage of the cpu-nb to stabilize, you may or may not need to increase the HT speeds/voltages too, stock is no longer fast enough maintain the communication between your ram and cpu. You may (or may not) need to bump voltages to your ram (I needed to bump the voltages for my ram a little) to stabilize.
Because i found vcore and temps got a bit out of control over 3.8, when i brought it back down under 3.8, i tinkered with the overclock to get the best temps/voltages. In order to bring down my vcore and general temps i switched to 100% FSB overclocking... while my cpu didn't love FSB overclocking i was able to knock 0.1v and nearly 10C off my cpu temps for an identical multiplier overclock. Something to think about as you go forward.