I prefer a simpler approach:

s/$c$a/$c$b/g;

or

s/($c)$a/$1$b/g;
[download]

</Editor's Note>

All right... here's a benchmark of the two methods (lookbehind and just straight substitution). It looks like lookbehind is slightly faster, but not by a huge lot. Here's the code:

    use Benchmark;
    use vars qw/$a $b $c/;
    ($a, $b, $c) = qw/bar baz foo/;

    timethese(shift || 1, {
        lookbehind => sub { local $_ = "foobarquux"; s/(?<=$c)$a/$b/g 
+},
        chromatic  => sub { local $_ = "foobarquux"; s/($c)$a/$1$b/g  
+},
    });
[download]

    Benchmark: timing 500000 iterations of chromatic, lookbehind...
    chromatic: 15 wallclock secs (14.46 usr +  0.00 sys = 14.46 CPU)
    lookbehind: 13 wallclock secs (11.93 usr +  0.00 sys = 11.93 CPU)
[download]