So far, the problem looks like this:
Because the box has only 2 GB of RAM, and .LRN isn't the only thing running on it, there is over 1 GB of swap in use. It appears that Oracle's SGA resides at least partly in swap (looking at iostat to see lots of swap activity while queries are run in sqlplus). This, of course, just kills performance.
To make matters worse, most everything is installed out on the disk array, so all of the log files, both nsd and Oracle, are being written to over a single SCSI channel.
My recommendation is to first get some more RAM, at *least* bring it up to 4 GB, and then if possible split things up across multiple disks, either by moving the log files to an internal disk or attaching a second disk array.
I'm pretty sure things will be fine after that, but if not, we'll continue looking into it at that point. I've tried doing a little query tuning but it's a lost cause right now; nothing I do makes any difference.
