Hobby servos are generally not sufficient for real robotics for a number of reasons: Quality, precision, range of motion, torque, etc.
Meanwhile, industrial servos, such as ABB, Emerson, GE, etc, are generally both heavy and expensive, and not suitable for small-humanoid scale actuation. Similarly, building your own servo from motors, gearboxes, and encoders, is akin to trying to design your own CPU just to control a motor -- too much detail getting in the way of real work.
There exists an in-between level of servos -- reasonably priced, reasonable performance, and reasonably controllable -- in the form of the competing brands of Dynamixel and HerculeX servos.
The smallest offerings in those lines generally are not strong enough for real-world interaction, but the next step up hold a lot of promise. For the Robotis Dynamixel line, this is the RX-24F servo (priced between the cheap AX-12F and the next step up the MX-28R.) Asking around, it seems that the specs and interface on that servo is great, but that it shuts down from thermal overload if you actually try to run it at rated load -- something that I'd expect from a hobby servo, but not a robotics servo.
Now, stepping up to the MX-28R doubles the price. Thus, if the RX-24F heat flaw could be fixed, it would be positioned at a nice price/performance point.
Does anyone have experience in providing additional cooling for this servo? Anything from thermal-gluing heat sinks to the case, to drilling holes and running cooling fluid tubing to any hot parts on the interior would be reasonable approaches. However, before I spend significant time and effort investigating this, I'd like a second opinion -- is it possible, does anyone have experience doing this, is it worth it?