Hi,
Well after blowing up a couple of servos due to a bad battery, I have built some simple code to mimic a walk with 3 legs on my Hexapod (new Servos are on the way). The walk is a simple 3 legged sweep forward then back, in a 20 count loop. All was I looking for was SD21 function and to see if I had any major position issues. At 6 volts positioning was not always good, at 5 Volts much better.
I did notice on my servos they got a little louder as the loop progressed, this is most likely a result of servo quality. It seems to be only a couple of servos so no big issue there.
The Pod is sitting on a paint can so no weight on the servos at the moment.
In the process I discovered a few things that you may l already know, but just in case.
The cheap servos are very voltage fussy, I found if the voltage was not steady and within the middle of the range, the positioning was quite poor. To clean up my voltage I installed a BEC, Battery Eliminator Circuit. In the R/C flight world it allows you to eliminate the battery for the receiver by supplying a steady 5 volts from your primary battery, and is ideal for small robots using this board.
My current BEC is only 10 Amps, but I have a 20 Amp on on the way. This allows me to provide a steady 5 Volts to the Servos, the 20 Amp is adjustable for output voltage. The down side is you need a fan, these generate a fair bit of heat under load and are designed to have airflow over them. I have a little 5 volt computer fan I run off servo connector 21, it is about an inch square and generates enough airflow to keep things cool.
I found by using the BEC, I both protect my servos and seem to get better positioning, the power supplied to the servos side is only running the servos. Currently I use a 5 volt run from my bread board to power the logic side. It is just a little simpler for debugging. But it may also assist in keeping things cleaner. When the bigger BEC comes I am going to running them both off the main battery and see if there is any difference.
I also found better positioning by slowing down the servo movement, I pass a speed, currently 12 (randomly selected). Although it is a bit slow the servos seem to come back the same position. At least they have not knocked over my endpoint sticks a small straw stood up :).
I need to play with that value.
Overall I am finding the SD21 does the job, but information is hard to come by as you noted it does not seem very popular.
My final positioning tests will happen about week from now, I have a small FSR and have come up with a rig I can position under a leg. In theory if the leg returns to the same position each time the FSR should have about the same reading. I figure about 100 iterations at a variety of speeds should test the servo positioning pretty well and give me a set of benchmarks.
I use a Castle Creations BEC you can find some info here http://www.castlecreations.com/products/ccbec.html I happen to be familiar with them, there are likely similar solutions in the robotics world.
I also switched to a different I2C library, primarily for better error control but it has some, at least to me nice features.