Fred, CCDStack is an exception. Stan Moore has produced a remarkable piece of code when it comes to noise suppression. The Sum algorithm does absolutely nothing to control noise. As you combine noisy images, not only does your signal increase, but so does noise at the same rate. This is where CCDStack smarts come in to play by performing significant analysis before the combine routine. I feel certain that using median or sigma with CCDStack would deliver an improved result. I'm not sure if you've tried Sigma by Ray Gralak -
http://www.gralak.com/Sigma/. Ray is also the developer of PEMPro. The code works quite well, but I personally like Russ Croman's RC Console for Sigma as it give you a nice percentage of how many pixels have been rejected (similar to CCDStack). While I do have a copy of CCDStack, I have not used it enough to make a quantitative comment on its different combine methodologies.
The common thread along all algorithms is large quantities of data (as you are well aware). These algorithms will not magically turn your average astro image into an APOD candidate. Only when you present the algorithms with such volumes of data will they work well for you.