FWIW, note that in the second example here, the outer for-loop is not needed:

c:\@Work\Perl\monks>perl -wMstrict -le
"my @re_subs = (
     sub { s/(?:this|or|that)/bob/gsi  for @_ },
     sub { s/(?:One|Two|Three)/Four/gs for @_ },
 );
 ;;
 my @strings = ('tHiS one or ThAt other', 'One two Three');
 printf qq{'$_'  } for @strings;  print '';
 ;;
 $_->(@strings) for @re_subs;
 printf qq{'$_'  } for @strings;  print '';
"
'tHiS one or ThAt other'  'One two Three'
'bob one bob bob other'  'Four two Four'
[download]

I believe that we're mixing terminology here, as "pre-compile" isn't exactly what you're wanting to ask. Yes, the code is only compiled once and after that point it will not need to be recompiled.

I've used this type of mechanism to program configurable filters. Letting a user edit a config file where they specify anonymous subs like that which are later eval'd and used in my larger package.

my $sub =eval "sub { s/$_->[0]/$_->[1]/$_->[2] for (\@_)}";
[download]

In an eval evironment in which  @_ was, e.g.,
    @_ = ('searchPattern', 'replacementString', 'regexModifiers');
(presumably, the eval statement is called in the context of a function to which arguments were passed via @_), the eval would return a code reference equivalent to
    sub { s/searchPattern/replacementString/regexModifiers for @_ }
which would allow one to iterate over a list of strings and do substitutions on each (non-literal!) string:

my $x = 'some';
my $y = 'strings';
my $z = 'here';

$sub->($x, $y, $z);
[download]

The only benefit that I can see of using the string eval is that the regex | substitution operator modifiers  /g /e /r and perhaps a couple others cannot be "passed" in any other way. If not for these modifiers, one could simply write something like (again, assuming this is in a function in which  @_ held search/replace strings)
    my $sub = sub { s/$_->[0]/$_->[1]/xmsg for @_ };
    ...
    $sub->($x, $y, $z);
(note the literal modifier group) and get the same result. In this example, the modifiers other than  /g could be passed in a string.

Update: E.g.,

c:\@Work\Perl\monks>perl -wMstrict -le
"S('(?xmsi) foo', 'bar', 'g');
 ;;
 sub S {
   my $sub = eval qq{ sub { s/$_[0]/$_[1]/$_[2] for \@_ } };
   ;;
   my $x = 'foo';
   my $y = 'Foo fOo foO';
   my $z = 'FOO';
   ;;
   $sub->($x, $y, $z);
   print qq{x '$x'  y '$y'  z '$z'};
   }
"
x 'bar'  y 'bar bar bar'  z 'bar'
[download]

And yes, this is a trick for a completely trusted environment. But then: "Trust, but verify!"

---

Give a man a fish:  <%-{-{-{-<

Speaking of evaling substitution into existence, if you trust the OP to specify the regex and substitution string, and you're not validating those strings

you might as well switch the interface/api to needing subs , so the user specifies the whole sub instead of parts to build one

on the other hand, String::Interpolate/String::Interpolate::RE

what do you observe happening when you remove eval?