Seems that this community is just immune to learning ....
A few thoughts:
Buying G5s, IBM-NBs, AMD-SBs and so on in small numbers (~1000) will be rather expensive, and who would pre-finace such a prodution run (we are talking about something ~300000$) ? Doing it in even smaller numbers will again increase the price per board even further.
And that assuming that IBM/AMD will even talk to you....
You just can't develop that kinda HW from your bedroom, you'll need lots of eququiment, experience and spare parts to replace those you burned in failed attempts.
Look at what time/effort Jeri/Jens put into the the C-One, a design based on really slow clockspeeds.
Look at how "far" Oli has gotten with his CF-upgrade, CPUs that are "all-in-1-designs", where one doesn't have to hassle with NB, SB and suchlike.
Look at how much time and money bPlan invested into the Pegasos1/2, based on technology 1 or 2 generations behind the G5, and with people who have some real experience in PPC-HW-design ....
_________________ - We don't need good ideas, we haven't run out on bad ones yet - blame Canada