One suggestion: if each record has more than 16 fields, you might find slightly better performance with

my @row = split /\t/, $_, 17;
[download]

which tells perl to split into at most 17 fields (0 to 15, leaving the trailing data in 16).