I tried the "modify the onboard regulator approach" and successfully booted a K6-III at 66 x 3, but could not get anything faster to boot. The power supply just could not handle the current, and the poor Q16 was getting quite hot. I tried the tinsnips / hobbyshop sheet copper / soldering gun heatsink construction method, and was able to run the chip at 200MHz for well over an hour.
Wanting more speed, I purchased the Maxim 1711 evaluation board from www.maximic.com for $60, and by adding a 1K resistor across C18, then turning all DIP switches to "short", I got a pure 2.2 v. I also needed to short the 5 ohm R12 to increase the amount of deliverable power.
Iopener Q16 was totally removed; Power for the board comes from the onboard 18v supply off the inductor side of Q14. Ground is ground, and the output goes to the lower Q16 pad. +5 bias from the eval board goes to the top side of iOpener R306. As a result, the power supply turns on when the IOpener turns off, goes off when it goes off, and pulls from the native 18v supply. No need for external power. I added a few more 1500 uF caps in parallel to the ceramic caps around the CPU.
The board will run at 4 x 100, but the N1 transistor gets hot if doing numerically intensive work. At 3.5 x 100 everything works fine.
I've ordered an additional IRF 7811 from Newark Electronics to parallel across N1 to improve drive / reduce heat. Hopefully this will let me go to 4.5 x 100.
Just wanted to let people know there is an alternative to riser sockets and external power supplies if you don't mind heavy soldering.