Test 1
Starting at 117 degrees, each program was enabled.  The program was given 7 minutes to cool the CPU as much as possible.  This test was performed on each program 3 times and the results were averaged.  The system was rebooted between each test, and the temp was allowed to reach 117 degrees before the program was run.
.
 
 

Not much difference here between the best and the worst, and even less difference between the majority. Test 1A
All of these programs do a great job of lowering the Idle temperature of the CPU.  The temperatures shown in the next chart are the result of letting the system sit idle for about 20 minutes.  The difference in cooler program temperatures are mostly due to the differences in room temperature, not the ultimate "cooling" ability of each program.  My computer room temperature is very hard to regulate exactly as the system and monitor are constantly dumping heat into the room and the A/C thermostat resides elsewhere.

The three cooling program's temperatures had stabilized well before the 20 minute time limit had expired, while using no program, the temperature continued to slowly rise, ultimately to 110.1° F.
.

I had initially intended to run each program through an entire suite of WinBench's benchmarks and note the temperature for each test as it ran.  After trying this for a few runs, it became apparent that there was too much room for error on my part.  If I didn't happen to catch the temperature at the same point in time for each test, the results would be less than accurate.  I ended up deciding just to note the performance figures  for this suite of tests and then re-test the individual components of the suite for both temperature and performance.

In the past, I had noticed that every once in a while I would get a set of results that were much lower or higher than the average results for a suite of WinBench tests.  I had thrown these results out and re-tested.  Most of the results came within 1% of each other and I used that as the criteria for those I kept.  I didn't think too much about the non- conforming results until I thought I started to notice a pattern. (I'll get to this a little later on.)

Each program was tested at least 3 times with the results averaged.  The hard disk was defragged between each test. The system was shut down and allowed to cool off  to 80°F then rebooted for each test.
.

* In the vast majority of tests, the results were as shown.  However CpuIdle 2.4 and 3.0 showed a higher incidence of greater than 1% fluctuations than either Waterfall or Rain.  The scores in a few tests went as low as the 630's.
.
Test 2A
The main thing I learned from testing with WinBench is that the cooler programs do little to degrade the system performance.  The results shown represent the best repeatable marks I could achieve with each program.  The temperature readings were achieved with the memory function of the thermometer and depict the highest temperatures that were reached during the tests.
.

This is where I noticed that, for some unknown reason, all versions of CpuIdle were more prone to producing a result that did not stay close to the average (except for version 3.01).  The ratio of non average results for CpuIdle ran about 1 (non-average) in 4 to 5 (average) attempts.

Using no program, Waterfall, and Rain, the ratio was in the 1 to 15 range. (The amount of the discrepancy was usually much less, also.)  At this point, I began to wonder if this behavior was caused by WinBench, CpuIdle, or the combination of the two.

For this test, the same defragging and rebooting procedures as before were followed.  I tried to bring the temperature of the processor back, as close as possible, to the same  reading.  Because it was not always possible to do this, starting temps are given.  The high, low, and average results are noted. (Average results do not include any results not within 1%.)  Each test was run at least 5 times, many more in the case of all versions of CpuIdle.
.

Test 3A CPUmark32
I felt that this test gave no new information and did not use it for this round of testing.
.

Test 4
With the hope of trying to track down the reason for the wayward CpuIdle processor results, I decided to try running Wintune97 from Windows Magazine.  I haven't run this benchmark much lately because it doesn't seem to stress the system enough to show small changes in system configuration.  It didn't change my views any with these tests.  Out of the 30 times I ran it (ten programs, three times each), all of the results for the Dhrystone and Whetstone tests were identical except for the three tests of CpuIdle 2.52b, which scored four points below the rest of the group in Dhrystone.  The only reason I chose to show this chart at all was the temperature readings for CpuIdle 2.4, 3.0, and 3.01.  These three programs seem to allow the processor to heat up any time the hard disk is accessed.  There will me more information on this in tests 6 through 8.

All of the usual defragging and rebooting procedures were followed.
.

Test 5
Still trying to track down the errant CpuIdle results, I turned to a benchmark I had heard good things about, but hadn't yet tried.  Bench32 from U Software. (defunct url, see below)  The shareware version is limited to a CPU test, however this was exactly what I needed to check the WinBench98/CpuIdle results.  Comparing one version of a benchmark to another from the same company is usually unreliable.  Trying to compare the results of two completely different benchmarks is total futility.  However, my purpose was not to compare benchmarks, but to see if I received any non-average results.  I didn't.  I have to conclude that the problem either lies with CpuIdle, WinBench98, or the interaction between the two.

I ran CpuIdle 2.4, 3.0, and 3.01 a total of 10 times each to check for non-average results.  The remaining programs were run 3 times each.  Results were averaged.  The system was rebooted between each test.
.

.

The hard disk was defragged and the CPU was allowed to cool between each test.  Each test was run 3 times and the results were averaged.

Test 6 CD to Hard Disk - K6

Test 6A
I have expanded the temperature information to include both beginning and ending temperatures.  By the way, the transfer times for tests 6A, 7A, and 8A were within a couple of seconds of each other in each category.
.

The same test procedures as before apply.

Test 7 Hard Disk to Hard Disk - K6

Test 7A
For these tests, I raised the file size to 500 MB which is equal to just under two minutes of continuous transfer time.  These results also give an indication of the temperatures reached during disk defragmenting, except that disk defragging generally takes longer and the temperatures rise accordingly.
.

Test 8 Hard Disk to Floppy Disk - K6

Memory Used   Chart 3

CpuIdle's "load only" and "control panel" modes take the honors for least memory used.  Also of note here is the fact that Waterfall Pro has added many new features while decreasing its non-shared memory and only adding 120 kb to the total usage.   Impressive.

n/a = not applicable or no results shown
Bold indicates programs tested in "A" AMD tests.
.
 Conclusion
No matter which of these programs you choose to use, you will benefit from a cooler running processor.  With the exception of the file transfer tests, the results are very close and it is doubtful that you would notice a difference in performance between the programs.  Please keep in mind that the results will vary with the processor type, system, and configuration used.  I have not tested these programs on any systems but my own.  Your mileage may vary.

Which program do I use?  I run Cpuidle 2.4 on my overclocked Intel 486-100 where heat is very much an issue due to the close confines of the case and the fact that the processor is a 5 volt overdrive unit.  Rain 1.0 handles the heat on the extremely hot, overclocked AMD K6, and have just switched to Waterfall 123 for the mildly hot, overclocked Celery.  How's that for covering all of the bases?

Updates will be added as they occur.