Your output sample shows employee 3379 twice. I'll assume that is an error.

You are assuming you already have a list @inlst of employees, though your code doesn't show where it comes from. The program should probably build the list as it goes along.

Here is one way to do it (__DATA__ section not shown):

use List::Util qw( sum);

my (@emp_lst, %prod);
while ( <DATA> ) {
    my ( $emp) = m{<emp>(\d+)</emp>} or
        warn "No <emp> in line $.\n", next;
    push @emp_lst, $emp unless $prod{ $emp};
    $prod{ $emp} ||= [];
    my ( $prod) = m{<prod>([.\d]+)</prod>} or next;
    push @{ $prod{ $emp} }, $prod;
}
for ( values %prod ) {
    $_ = @$_ ? sum( @$_)/@$_ : '';
}
print "$_ - $prod{ $_}\n" for @emp_lst;
[download]

Anno

Update: I agree with GrandFather that an appropriate parser is a better solution.

Update: Noticed an optimisation. The pesky $prod{ $emp} ||= []; can go from the while loop and everything rely on autovivification as it should:

while ( <DATA> ) {
    my ( $emp) = m{<emp>(\d+)</emp>} or
        warn "No <emp> in line $.\n", next;
    push @emp_lst, $emp unless $prod{ $emp};
    push @{ $prod{ $emp} }, m{<prod>([.\d]+)</prod>};
}
[download]