Thanks for the replies.
Sadly the board revision doesn't support any coppermines so I'm limited to the 250nm Katmai ones - 600, 550, 500, 450. TDPs are roughly 25 to 35. Presumably I could run the 600 at a lower multiplier and have it at the TDP of the 450, if I was to go P3, giving me some flexibility should I ever need it. I'm not sure how low that could be set.
If I avoid the 350nm Klamath P2s then TDPs are broadly in line with the P3s based on clock speed. This would allow a P2 450 down to a P2 266 (the 233 is a Klamath part).
So I guess that leaves me with the following questions, which I don't know can be answered without testing.
Will P2s work at whatever downclocked multiplier and FSB they are set to, or do they need to match?
Will P3s work at whatever downclocked multiplier and FSB they are set to, or do they need to match?
And if downclocking works does it need to be a valid CPU speed that I match, or could I get a P2/P3 working at 66x3=200mhz, which is the lowest combination the motherboard DIP switches allow.
It would be perfect if that were the case, I could put a 600Mhz P3 in and have it run at a much lower wattage forever, knowing that if I do encounter a game that I just have to get running in Win98 or DOS and not XP I have some extra performance I can use.
Edit : Some searching seems to indicate that Klamath is the last one that can be changed. I knew P3s onwards could not be overclocked, but it seems they cannot be underclocked either.
This means that I get a Klamath which defeats the whole purpose, or I am limited to underclocking them by reducing the FSB only.
More edits : Further checking on the power consumption shows that it scales more or less uniform based in relation to CPU speed, so a 600mhz operating at 66FSB instead of 100 should operate mostly as a Celeron 400 (cache aside) and power consumption should be the same.