Well, there's always that middle ground between C and perl - XS modules. Cut probably still beats it, but Text::CSV_XS appears to be faster than split for files of that size. And has the added advantage of supporting embedded newlines and embedded delimiters.

For a 1 million row table of 100 columns, sum the second column:

             s/iter        Split Text::CSV_XS
Split          49.8           --          -8%
Text::CSV_XS   45.9           8%           --