The noise level of an MDD is, unsurprisingly, an interdependent amalgam of the contributions from CPU (heat output), 120x120 fan speed, throughput of air and noise (cooling effort), 60x60 PSU-cooling fans which I misdescribed as 40x40 above (noise), heatsink (cooling effort), thermal paste composition (cooling effort), AppleFan.kext (fan speed control), and Napmode script (CPU duty cycle and heat output), the sum effect of which can be shown by Temperature Monitor, Activity Monitor and Hardware 'control panel' (from CHUD) in System Preferences.
My dual (now 1.42GHz) G4 MDD settles during moderate activity (as shown by Activity Monitor) at about 25°C in an ambient of 16-18° (winter) or, currently, 30° in an ambient of 24° (spring). When it was a dual 1.0GHz with aluminium heatsink and no-name thermal paste it reached over 50° regularly with the same moderate activity in an ambient of 30°.
I modified the .kext after
http://www.s155158671.websitehome.co.uk/applefankextmod.html and
http://homepage.mac.com/paul74/MDDFanExtensionMod.htm
but lowered the range with the same intervals. I downloaded Apple's CHUD 3.5.2 for the CP Hardware, which I use with Tiger and Leopard on different HDDs in the same MDD. Hardware misreports the L2 and L3 cache sizes under Leopard, but does not disable them. The CHUD script NapMode (for which there is a mod. to use under Leopard) is discussed at
http://www.xlr8yourmac.com/feedback/MDDCHUDfeedback.htm
and Temperature Monitor is found here
http://www.bresink.com/osx/0TemperatureMonitor/screenshots.html
I use a copper heatsink over a minuscule film or 'smear' of Arctic Silver (enough only to fill the microscopic irregularities between the processor die and the heatsink). Treatment of the PSU's fans is mentioned in the post above. The fan for the heatsink is the original Papst fan. Inherent differences between AcBel and Samsung PSUs are, I suspect, overwhelmed by the modifications to software and hardware. Made carefully, the modifications work well.