The benchmark result is confirmed for better performance of substr in the test above.
I've put $count = 100 and reduced the input file size. Now I don't get any warning, the improvement with method Three is evaluated to be about 25%:

Benchmark: timing 100 iterations of Method One, Method Three, Method T
+wo...
Method One:  7 wallclock secs ( 7.60 usr +  0.11 sys =  7.71 CPU)
Method Two:  8 wallclock secs ( 7.83 usr +  0.11 sys =  7.94 CPU)
Method Three:  6 wallclock secs ( 5.82 usr +  0.10 sys =  5.92 CPU)
[download]

About the benchmark... I felt more confortable with about 40 CPU seconds, even if there was a warning. With repeated iterations I think some time is taken by the call, while one call with a big file is better focussing on the array extraction. But it is just my feeling... :) I'd better study the Benchmark routines to evaluate this
ciao
Roberto