The standard method for calculating words for the purpose of finding WPM is counting all the characters and dividing by six. Why six? Because the average word is 5 characters long, plus a space.

I'm really no good at one-liners or golfing, but I quickly modified your code. It obtains the accuracy you're looking for. Tested on ActivePerl Build 629 (Perl v5.6.1).

#!perl
$x=time;
$letters=scalar(split//,<>)/6;
print $letters/(time-$x)*60;
[download]

golfed .... with line breaks and prefix

perl -e'$x=time;print "WPM:".(scalar(split//,<>)/6)/(time-$x)*60."\n"'
[download]

subtract ."\n" and add e (thanks once again to chipmunk)

perl -le'$x=time;print "WPM:".(scalar(split//,<>)/6)/(time-$x)*60'
[download]

use Time::HiRes qw(time);
[download]

This will give you a floating point time function instead of integer seconds.

In the current program you will have to type a really long line so that the division errors like 4.9 seconds which will sometimes be 4 seconds and sometimes be 5 seconds which will cause a 25% change in your typing WPM.

The floating poing time function will fix this up since the steps will be on the order of 0.01 seconds rather than 1 second which will improve the granularity of your estimate 100 times so that a one step change from 4.99 to 5.00 will result in a 0.25% change in the WPM figure.