Rather than speculate on efficiency you can always use benchmark. One small point though - whenever you interpolate a string into a m// regex or the first half of a s/// regex you need to backslash your special regex metacharacters $^*()+{[\|.?

The easiest way is to use quotemeta.

foreach $key (keys %hash) {
    $key = quotemeta $key;
    $text =~ s/$key/$hash{$key}/eg;
}
[download]

This is *vital* for reliability. Otherwise you will get unexpected runtime failures when your data eventually contains metachars (typos, deliberate, malicious...)

use Benchmark;

timethese(10000, {

    'Simple loop' => '

    $text = "This is my test string";
    %hash = qw(test text foo bar use loop the end);
    foreach $key (keys %hash) {
        $key = quotemeta $key;
        $text =~ s/$key/$hash{$key}/eg;
    }

                     ',

    'Alternation' => '

    $text = "This is my test string";
    %hash = qw(test text foo bar use loop the end);
    $regex = join("|", map(quotemeta, keys %hash));
    $text =~ s/\b($regex)\b/$hash{$1}/eg;

                     ',
    }
);

Output:

Benchmark: timing 100000 iterations of Alternation, Simple loop...
Alternation:  5 wallclock secs ( 4.12 usr +  0.00 sys =  4.12 CPU) @ 2
+4271.84/s (n=100000)
Simple loop: 10 wallclock secs ( 9.72 usr +  0.00 sys =  9.72 CPU) @ 1
+0288.07/s (n=100000)
[download]

So as it happens alternation is twice as fast. I thought your solution would be faster but there you go!

If in doubt - use Benchmark

Cheers

tachyon