When you OC by upping the fsb, it also raises the speed that your peripherals run at. Many components have difficulty in running at this higher speed which causes stability problems on that system.
As for the heatsink question, I am putting a Volcano II in my new system and if I am not satisfied with the temps I will probably get an FOP 32-1 or something like that. The reason people run "nice coolers" on their non OC'd processors is because cooler = more stable even if the chip is not OC'd. In the case of the AMD TBirds, they run very hot to start with so good cooling is important right from the beginning. Hope this answered some of your questions.
What greatmatt says about increasing the FSB also increasing the PCI bus is mostly true. If you O/C a 100 MHz FSB CPU to 133 MHz however, the 1/4 PCI divider kicks in and returns the PCI to the spec frequency of 33 MHz.
Also, I have never had any PCI card that was bothered by a higher PCI clock. In my old system a couple of years back, I O/C'ed a slot 1 Celeron 366 to 458 MHz by upping the FSB to 83 MHz. This set the PCI at 41.5 MHz, way out of spec. But, it didn't bother either of the two sound cards or one PCI video card I used. It did, however, screw up some old Fujitsu and Maxtor hard drives, but I picked up a Western Digital that ran fine in the system for over a year. Today most hard drives can run at higher PCI clocks though, so it's not as big a problem as it was a couple of years ago.
I know there are some cards, particularly NICs, that are sensitive to higher PCI clocks. But I don't think that "many" PCI cards would be affected by increasing your FSB. And increasing the FSB improves total system performance better than simply playing with CPU multipliers since everything in the box runs faster, not just the CPU.
As far as CPU H/S fans go, AMD Durons and T-Birds run very hot and need good cooling even to run at default. You should at least get the GlobalWin FOP32-1 that greatmatt mantioned - it's only about $20 and does a good job. For a non-overclocked Celeron or PIII it's OK to go cheap as they run pretty cool.
In reference to what you said Dave regarding PCI bus acceleration My concern actually is related to what was likely the problem with your HDs
I have several machines that i run together (synched) that have digital audio being pumped through cards in each one. The cards are about 3 years old and are 33mhz cards. If I accelerate the bus on the systems (Synonymous with increasing the bandwidth)and get the bus out of a multiple of 33, I wonder if it is going to cause problems with the audio cards because they are basically full time data pumps that are constantly receiving and sending large amounts of data like hard drives. The cards are PCI 2.1 compliant , and I know that means they can exist in a 66mhz environment but i don't know if that means that they can exist in a bus that is not a 33 multiple. This is likely the problem with your older drives which might have been designed to a bus spec which was quite some time back. Also, I really don't know the danger implications for the cards. As they were about a grand apiece, I don't want to risk frying them if that is an option, but I would really like to have the system performance of OC if I could try it without ultimately harming them even if it didn't work.
Do you guys know whether there is danger for cards to put them into an accelerated bus if they can't work in it?
Also, what do the various PCI spec compliances indicate for a piece of hardware in the new systems (I know that we're on 2.2 now)?
Dunno what to say about 3-year old PCI cards. But being a multiple of 33 isn't an issue. The issue is whether the individual gates in the chip transistors can switch states at the clock rate. If they can, it'll work. If not, they won't. I don't know why they would "fry" just because of a little overclocking.
About the hard drives, there was no physical damage. My sister-in-law is still using the Fujitsu which corrupted horribly. Only the data was lost, and both the hard drives were both fine after being re-formatted. The use of DMA also contributed to the corruption. I now have a 7200 rpm Maxtor, which isn't troubled by my overclocked PCI bus. I've only had it up to 38.75 MHz (155/4) in my current system though.