Performance enhancements for single and multi-thread evaluations

[uw4]

Experiments show that with few changes huge boosts (synthetic tests 3x faster) can be achieved
(caused by less synchronization + much better JIT code generation)

[uw4]

Perf enhancements speedup:

Before:
$ ./threads.sh 1
Testing THREADS=1 TIME=3000
Finished 17 tests sum=2.605197E11
Thread test finished total=17

$ ./threads.sh 4
Testing THREADS=4 TIME=3000
Finished 12 tests sum=1.689628E11
Finished 12 tests sum=1.929169E11
Finished 12 tests sum=2.02259E11
Finished 11 tests sum=1.705385E11
Thread test finished total=47

After:
$ ./threads.sh 1
Testing THREADS=1 TIME=3000
Finished 42 tests sum=6.101973E11
Thread test finished total=42

$ ./threads.sh 4
Testing THREADS=4 TIME=3000
Finished 26 tests sum=3.680051E11
Finished 22 tests sum=3.780026E11
Finished 26 tests sum=4.324088E11
Finished 27 tests sum=3.765051E11
Thread test finished total=101

[aeberhart]

maybe update the performance numbers in the README?