A Data Science Central Community
I believe that Mahout is a bit weak on Linear Regression, and does not use a logit model for logistic regression. I think it is uses another optimized iterative learning algorithm, rather than maximum likelihood.
A good person to pose this question to would be Sean Owen, who is very active on Quora.
Check out the RMS package in R (http://cran.r-project.org/web/packages/rms/index.html). The author, Frank Harrell, is a renowned Biostatiscian who has been programming stats routines for many years.
One assumption that Mahout may be making is that with big data virtually all variables will be significant so statistical significance is not the best measure here.