Performance is now looking quite good - 2.4 s avg

[Patrick Nichols <pnichols@graphics.csail.mit.edu>]

All --

Some more aggressive caching and a bit of profiling has reduced average 
query time a bit more. The lastest results:

** Completed    200 trials
** Mean Time:   2401 ms
** Avg. Spaces: 21196

So, from 300s to 2.5s is a pretty good improvement. I tried switching to a 
float implementation, but this didn't make a substantial difference -- so 
I suspect either (a) double ops are somehow implemented in hardware, or 
(b) floats are actually implemented as doubles in java, under the hood. 
(Anyone know about this?) Either way, these numbers are almost good 
enough. I'll look into super-node sampling to see where likely outliers 
may be as soon as I get some data from MIC/Vitaly.

-Patrick

-- 
* * *

Patrick James Nichols II
Graduate Student, MIT Computer Graphics Group
http://graphics.csail.mit.edu/~pnichols