Eradicatore: I want to read out say 1000 bytes of a file,
and then scan that 100 bytes for a string.
But I do NOT want to first unpack it because
I'm hoping that avoiding the unpack may save some time.

OK, ikegami answered to that already. I'd like to add that
index() would be the fastest pure-Perl solution. If its a
very big binary chunk, write a small "Inline => C {}" wrapper
to C's (stdlib) memchr() function. This might be, depending on the
architecture, up to three times faster than index() (on large chunks).

If I do unpack, what's the best way to do that?

Your solution would be o.k., you might consider to do
a pseudo-Schwartzian to map the indices into your target
array @tmp, sth. like:

open my $fh, '<', 'test.elf' or die "can't do anything: $!";
binmode $fh;
read $fh, my $stuff, 1000 or die "read error: $!";
close $fh;

print length $stuff, " bytes in\n";
my $offs = 0;

my @tmp =
   map $_->[1],
      grep $_->[0] eq '7f',
         map [$_, $offs++],
            unpack "(H2)*", $stuff;

# prints "7f" offsets in binary file
print join':', @tmp;
[download]

Regards

mwa