In response to my own question, above, I implemented a fairly powerful IFI co-processor that uses the debug port. I found out that by carefull processing you can transfer data between the Basic Stamp and an outboard processor at 28.8kb - fast enough to transfer 33/45 possible bytes (26 inputs and 18 outputs + LEDS) while maintaining the IFI 26ms cycle time. You could transfer all 45 variables, but would extend the cycle time to 35ms or so. The entire project can be seen here
http://www.barello.net/FIRST Look at the last two items.
Note, I don't have the board made yet. The plan is for 8 channels of 10bit high speed ADC and 16 bits if general I/O (user defined in code). The current code base has 3 channels of Quadrature decoding (for wheel position, motor velocity, whatever: It keeps up with 20krpm and a four slot encoder) Gyro code that tracks absolute theta from the start and a second serial port for debugging.
The actual board is on hold for the next month or two until I find out what IFI has in store for us (they are re-vamping the RC, apparently).
Cheers!