The RAM option has been something that has been discussed in length in the past. ALthough its certainly of interest, I think Howard just feels its not worth the time invested to have a scalable option for memory usage at this time.
Extra RAM usage wouldn't make a difference anyways. When you flip the extra memory switch, it loads itself almost entirely into memory (up to around 125-150MB in size generally speaking) and the amount that is stored on the hard drive is very little. Because it is kept in memory in runs faster because it doesn't read out to the hard drive, wait for the hard drive to find it, have the hard drive return it. ALthough its fast, relatively speaking its a lot slower than having it all in memory where latency is at a bare minimum. Short of a RAM drive, it won't entirely be in memory, but that's entirely different issue. The total amount of memory does vary with the protein size and I'm assuming it will be similar for Phase II client. On some of the smaller proteins, I've seen it take up less than 100MB of memory. Remember that the memory is shared so DF will adjust itself accordingly. If you have more RAM, than DF's memory usage will stay at a more constant level although if you ever watched the memory usage there is a small range of fluctation in the range of a few megabytes as it works.
Best,
RuneStar½