After doing some extensive experiments using a K6-III+ running at 2V I need to warn users that even 300Mhz may be too fast due to Q16 heating.
I found some power MOSFET's with better specs than Q16. My bright idea was to solder these in parallel with Q16 and mount a heat sink on it. I figured if I could find a MOSFET with higher conductance at the same voltage as Q16 most of the heat would go to it rather than the real Q16 as its resistance would be lower at the same bias level. I was hoping to avoid lifting Q16, mounting a cooling resistor, and lowering the core voltage. My heat sink was around 2" square of 10 mil copper. I bypassed the overcurrent cut-out with a .06 ohm smt.
I fired it up and all seemed fine. I warped up to 450 Mhz and all seemed fine, then the power supply crowbar'd and refused to reboot. I opened up the box and found that my new Q16 was so hot the solder had melted and the heat sink shorted something out.
I decided to solder a copper heat sink to the solder tab at the top of Q16. Attempting to cool it through the plastic case is very ineffective, but you can solder directly to the tab at the top. My heat sink was 10 mil copper again, 1/4" wide and 1.5" long. I bent it in an "L" after soldering it so it would fit under the RF shield. I fired up the board and kept it at 300MHz (100X3)for 5 hours. Motherboard Monitor gave me steady-state temps of 180 degrees F running a continuous loop of Windows Media Play, full screen + full volume sound. The heat sink was too hot to touch. I had no way to get a direct reading, but it was higher than 180 degrees for sure.
Anyone running at 2V should be warned that all that heat is going right into the motherboard. I suggest making some attempt to add heat sinking and forced air flow. I am looking into small 25mm 5V fans at this time to mount at the top of the I-O.
Bottom line is that you are at risk without serious airflow AND heat sinking if you run at 2V. It appears that use of a cooling resitor and lifting Q16 and adding heat sinking to it is the only good way to date to get serious speed. No amount of bypassing will allow you to stay above 400 Mhz without reduced core voltage.
It looks to me like the real answer is something like a powerleap socket to fit in the I-O.