First i want to ask you what do you mean with a false impression of it's performance???

I wasn't questioning your timing method. My point was that using vec on full 32-bit numbers is directly equivalent to using pack, but slower. Unpacking with vec is much slower:

C:\test>p1
$packed = pack 'V*', 1 .. 1e6;;
cmpthese -5, {
    unpack => q[ my @nums = unpack 'V*', $packed ],
    unvec  => q[ my @nums; push @nums, vec( $packed, $_, 32 ) for 1 ..
+ 1e6 ],
};;
         Rate  unvec unpack
unvec  1.45/s     --   -45%
unpack 2.64/s    81%     --
[download]

But even that belies the real problem. With Elias Gamma, you have to pack/unpack strings of bits that cross byte boundaries, and the only way to do that it to do them 1-bit at a time. That means calling vec for every bit in the string, rather than every 32-bits as you do in your test above. And once you start calling vec for every bit, things slow down dramatically. As you would expect as you are calling the built-in 32 times more frequently. By way of comparison, the equivalent of your test above, but done one bit at a time is:

use strict;

my $wektor = '';
for my $num ( 0 .. 9_999_999 ) {
    my $binNum = pack 'V', $num;
    vec( $wektor, $num * 32 + $_, 1 ) = vec( $binNum, $_, 1 ) for 1 ..
+ 32;
}
print "Vector's size: " .
  length( $wektor ) . " bytes\n";


my @vec;
my $Aa = time();

for my $num ( 0 .. 9_999_999 ) {
    my $binNum = '';
    vec( $binNum, $_, 1 ) = vec( $wektor, $num * 32 + $_, 1 ) for 1 ..
+ 32;
    push @vec, unpack 'V', $binNum;
}
print "unpack vector in \t", time() - $Aa,
    "  secs...(Oh dear!!!)\n";

__END__
C:\test>junk0
Vector's size: 40000001 bytes
unpack vector in        362  secs...(Oh dear!!!)
[download]

But, you still seem to be completely ignoring pack 'w*', @nums, which gives better compression than Elias Gamma, and is faster than anything else to boot.