After putting some stuff as Anonymous (#639826)
I did some research on the topic.

What's the status of the beer meter anyway?

OK, there is no way in pure Perl to come even
close to a tailored inline solution, so if its
important one has to use it.
I compiled several snippets from this thread
into one benchmark ==> http://hobbit.chemie.uni-halle.de/project/meterofbeer/
and added some results.

One very interesting outcome for me was the
revelation of how dead slow the Core2 architecture
was at repeated assembly string opcodes
like   repne scasb. By rewriting the scasb by a sequence of "mov"
it will be blazingly fast on a Core2.

There seems to be some error in the
sub "corion" ==>http://hobbit.chemie.uni-halle.de/project/meterofbeer/beerbench.pl
maybe somebody can fix it.

This is on a Core2/Q6600@3GHz (more results in the other link):

             Rate
split1     7.35/s
ikegami1   36.6/s
substr1    45.9/s
mrm_6      1169/s
avar2      1536/s
corion     1662/s
avar2_pos  2701/s
mrm_3      2819/s
ikegami2   3137/s
bart       3480/s
ikegami3   3488/s
mrm_1      3508/s
ikegami4   3655/s
moritz     3719/s
mrm_5      4271/s
mrm_4      4346/s
rep_scasb  4495/s
inline_c  11310/s
cmp_movb  11563/s

Interesting stuff!

Thanks & bye

Mirco