I'm curious, as I understand that answer this way:
If you are using a Linux derivative or something like cygwin on Windows, then this is an efficient solution for reading only the third field from your files ( assumes whitespace separated data, other delimiters can be specified ).
data= ReadList["!head --lines=-2 " <>#<> " | tail --lines=+39 | cut -f 3 "]& /@ FileNames
If one deals with data files in the 2 digit GB range, and often only some columns of the data are needed for further evaluation in experimental science, should one think about running mathematica on Linux, cygwin,... Is the efficiency boost for importing data small, big, huge? Do I have more options to select specific data elements in files?