XPS max I/OMB per CPU second

Question: Have you ever seen your system run XPS simple table scans at higher than 5-10 I/O MB per CPU second? Background: One common approach to sizing a system for a given workload is to try to balance the "horsepower" of components (processors, system bus, networks, HBAs, disks, etc) to avoid spending money on pieces that will never be stressed. MB/CPUsecond is a useful metric for characterizing the relationship between the I/O subsystem throughput and the processor utilization. For a given workload (simple scan, index create, table load, etc.), it gives a rough estimate of the most that the processors will ever ask the I/O subsystem to do. For example, say you're running XPS on six nodes, four processors per node with 'average' processor busy = 33% and the total I/O throughput is 80 MB/second. The CPU-normalized rate would be 80 / (6 * 4 * 0.33) = 10 MB/CPUsecond. Given linear scaling ;-); these 24 processors would never ask for more than 240 MB/second. I've watched this on SP2 racks, Sun E4k - E10k, Digital (CPQ) Alpha 8400s and HP K5xx systems and for simple table scans the range has always been between 5 and 10 MB/CPU second. Note that ratio this plummets for hairy joins or lots of aggregate functions or big sorts or lots of other things but I'm mostly interested in the highest value. I've only seen ratios this high for simple scans and maybe long checkpoints. So, the question is: Have you ever seen XPS run scans (or anything else) at higher than 5-10 I/O MB per CPU second on any Unix (or NT) system or is this the current state of equilibrium? Thanks, Doug