I just bought some new RAM for my G4, bumping it up to 2GB from 768MB - a substantial increase that I thought would have a decent impact on performance. But after installing the new RAM running CPU benchmarks, the floating point and integer speeds are just about exactly the same. System Profiler is seeing the RAM, and it's being reflected in my host profiles, so I'm not sure what's going on. Suggestions, anyone? Thanks in advance.

Extra RAM will not make any difference on the the floating point and integer speeds.
Only a faster CPU will make those numbers increase.

The additional RAM will help your projects run smoother and require less in-out swapping of virtual memory to/from your hard disk.

So is there any actual increase in crunching speed overall from adding more RAM? If not, what's the point?

Also, is there an optimum setting for "Write to disk at most every XX seconds" in general prefs that will improve overall performance?

If the L2 is not large enough, it causes swapping with RAM.

If the RAM is not large enough, it causes swapping with Disk.

Each step slows down the crunching, significantly. Having not enough RAM will slow you down. Having surplus RAM will not speed you up.

In your case, if you weren't already using 100% of your RAM, adding more will not help.

Adding more RAM will allow you to crunch for projects that have a minimum RAM requirement. For example, Predictor has required 1gb of RAM in the past.

That all makes sense - thanks for the info. Does the "Write to disk at most every XX seconds" setting have any impact on crunching speed?

If I understand correctly, the answer is "no". "Write to disk" could be better phrased as "save work every...". So it won't impact speed, except for the little bit of processing it takes to save a snapshot of work done to the HD. Less frequently will have a slightly better impact to speed, at the expense of a slightly longer loss of work if there is (for example) a power failure.