I get autobox to be slower as well, but not that much slower. Using a wrapper program to run your programs 10 times, and calculating a confidence interval, I get:

standard: mean = 3.881; 90% confidence interval: 3.809-3.952
 autobox: mean = 4.915; 90% confidence interval: 4.365-5.465
[download]

Another run, now with 95% confidence:

standard: mean = 4.100; 95% confidence interval: 3.866-4.335
 autobox: mean = 4.695; 95% confidence interval: 4.402-4.989
[download]