My Bounding Box timings (on dpmlh045).  Please note that I in no way 
expect you to get close to this, for this simple exercise.  I simply 
want you to see what sort of performance benefit thinking a little 
about your implementation can give.



For debug (i.e., -p -pg options to g++):
-----------------------------------------
   (100,000 random rays) "Found 4000 intersection in 0.300000 sec"
 (1,000,000 random rays) "Found 40522 intersection in 3.040000 sec"
(10,000,000 random rays) "Found 404033 intersection in 24.809999 sec"

Profiling (Debug Version):
                main()          fRandom()      BBoxIntersect()    RandomSphereVector()
--------------------------------------------------------------------------------------
  100000*  |        0%     |      50.08%     |     12.52%       |    25.04%
 1000000** |    10.16%     |      60.97%     |      8.71%       |     5.81%
10000000   |     8.93%     |      62.41%     |      8.53%       |     6.08%
    * = Not running long enough to get statistically relevant samples
   ** = Not running long enough to get consistant results from run-to-run
-------------------------------------------------------------------------------------




For optimized (i.e., -O3 options to g++):
-----------------------------------------
   (100,000 random rays) "Found 4000 intersection in 0.070000 sec"
 (1,000,000 random rays) "Found 40522 intersection in 0.720000 sec"
(10,000,000 random rays) "Found 404033 intersection in 5.810000 sec"