I wondered how a plain regex would fare. Probably not enough in it to be significant.

#!/usr/bin/perl
use strict;
use warnings;
use Benchmark qw(:all) ;

# Set up an array of data to test
my $data = '0906928472847292INR~UTRIR8709990166~     700000~INR~200806
+23~RC425484~IFSCSEND001                       ~Remiter Details ~10000
+07   ~TEST RTGS TRF7                     ~                           
+        ~                                   ~ ~RTGS~REVOSN OIL CORPOR
+ATION   ~IOCL  ~09065010889~0906501088900122INR~         7~         1
+~ 1';
my @data;
push @data, $data for 1 .. 1000;

cmpthese(
    1000, 
    {
        'Full split'      => sub {my @refnos; for (@data){my @splitted
+ = split(/~/,$_);   push(@refnos,$splitted[1]);}},
        'Limited split'   => sub {my @refnos; for (@data){my @splitted
+ = split(/~/,$_,3); push(@refnos,$splitted[1]);}},
        'Optimized split' => sub {my @refnos; for (@data){push @refnos
+, (split(/~/,$_,3))[1];}},
        'regex'           => sub {my @refnos; for (@data){push @refnos
+, /~([^~]+)~/;}},
    }
);
[download]

                  Rate    Full split Limited split Optimized split    
+     regex
Full split      23.3/s            --          -81%            -90%    
+      -91%
Limited split    120/s          413%            --            -51%    
+      -54%
Optimized split  244/s          946%          104%              --    
+       -7%
regex            262/s         1024%          119%              7%    
+        --
[download]