This will run faster, and IMHO improves upon split_join() a little ..

sub index_split_join {
    return $_[0] unless index $_[0], 'STOCK' >= 0;  # do a fast check 
+to see if line needs to be looked at
    my @tokens = split /\|/, $_[0];   # split into columns
    $tokens[15] =~ s/STOCK/BOXXE/;    # do replacement in col 16
    return join('|',@tokens);         # glue back together for final r
+esult
}
[download]

For your test of 1 data line, i get:

                 Rate    splitjoin idxsplitjoin simple_regex
splitjoin     50000/s           --         -15%         -60%
idxsplitjoin  58824/s          18%           --         -53%
simple_regex 125000/s         150%         112%           --
[download]

But that test isn't valid. Presumably (?!?) there are many lines that need to be processed, and only a small percentage have the word 'STOCK' in them (which is where the index short circuit will excel). Here is a modified benchmark (the DATA is ~1000 lines, all with same # of cols, but only a handful have STOCK in them):

my @lines = <DATA>;
cmpthese(10000, {
    idxsplitjoin    => sub {index_split_join($_) for @lines},
    splitjoin    => sub {split_join($_) for @lines},
    simple_regex => sub {simple_regex($_) for @lines},
});
# RESULTS:
Benchmark: timing 10000 iterations of idxsplitjoin, simple_regex, spli
+tjoin...
idxsplitjoin:  9 wallclock secs ( 9.16 usr +  0.00 sys =  9.16 CPU) @ 
+1091.70/s (n=10000)
simple_regex: 11 wallclock secs (10.77 usr +  0.00 sys = 10.77 CPU) @ 
+928.51/s (n=10000)
 splitjoin: 158 wallclock secs (158.15 usr +  0.00 sys = 158.15 CPU) @
+ 63.23/s (n=10000)
               Rate    splitjoin simple_regex idxsplitjoin
splitjoin    63.2/s           --         -93%         -94%
simple_regex  929/s        1368%           --         -15%
idxsplitjoin 1092/s        1627%          18%           --
[download]