Just couple of "performance numbers" from my testing for your comparison...
At begining, consider please, that my desktop is relatively ancient (INTEL two core @2.8 GHz, 2 GB RAM, 160 GB HDD).
I generated (STATISTICA 9, 32 bit) spreadsheet 100 VARS x one million CASES with double precision (8 bytes per cell). It lasts about 20 s and render about 800 MB temporary file.
Fulfilling (all 100 milions cells) with random numbers takes 1 min 01 sec.
Calculating mean, SD, and SE for all 100 variables: 30 s. Interestingly, each parameter calculated solely even together last always almost same time - half min.
Skewness and/or kurtosis calculation: 34s
Pearson correlation 100 X 100 (matrix from all variables against same): 49s
PCA: 1 min 23 s
CART: (first Var target, Var2-Var51 as continuous predictors, default parameters, Cross validation off):
5 min 18 s. Resulting tree has 91 nodes.
StatSoft patented "Feature Selection and Variable Screening" (Var 1 target, Var 2 - Var 100 predictors):
1 min 50 s.