Column explanation:
	Tasks: the number of "users" simulated, roughly a load factor
	Jobs/Min: the "throughput" of the system at this load
	JTI: a mathematical calculation of the fairness of the scheduler
	Real: amount of "wall" time this load took to complete
	CPU: amount of "CPU" time this load took to complete
	J/S/T: stands for "Jobs per Second per Task"
	user: reported user-space time in seconds
	sys: reported system (i.e. kernel) time in seconds
	idle/wait: reported "other" time in seconds (mostly disk wait)
The test stops when the "Tasks" column exceeds the "Jobs/Min" column - what AIM calls "crossover".  I didn't use the "adaptive" timer which finds the exact crossover, as that's not what I was looking for.

Linux 2.4.9
-----------
Benchmark	Version	Machine	Run Date
AIM Multiuser Benchmark - Suite VII	"1.0"	sundown2	Oct  2 07:47:40 2001

Tasks	Jobs/Min	JTI	Real	CPU	J/S/T	user	sys	idle/wait
1	11.8		100	492.0	10.9	0.1972	1.4	16.7	3917.9
500	2801.9		97	1038.6	7724.7	0.0934	701.0	7049.8	558.0
1000	1584.9		97	3672.2	20877.7	0.0264	1403.5	19525.2	8448.7
1500	977.3		98	8932.7	40021.1	0.0109	2118.2	37992.6	31350.9

Linux 2.4.10
------------
Benchmark	Version	Machine	Run Date
AIM Multiuser Benchmark - Suite VII	"1.0"	sundown2	Oct  1 15:19:53 2001

Tasks	Jobs/Min	JTI	Real	CPU	J/S/T	user	sys	idle/wait
1	45.3		100	128.3	4.2	0.7558	1.4	4.0	1021.4
500	9365.5		97	310.7	1983.8	0.3122	683.7	1301.3	500.7
1000	4250.0		98	1369.4	6144.7	0.0708	1384.7	4772.7	4798.0
1500	4022.7		98	2170.2	9567.2	0.0447	2077.3	7531.2	7752.9
2000	2832.5		98	4109.4	14865.7	0.0236	2773.5	12142.4	17959.6
2500	1881.5		98	7733.4	20815.3	0.0125	3482.7	17349.2	41035.0