I've made a try with a brand new install of cooking, installed firefox, created a list of all files belonging to firefox and libfirefox with tazpkg list-files (removing extralines with a grep).
If i made a backup with busybox's cpio, remove firefox and decompress the cpio with -id i have firefox back.
If i make the backup with the real cpio, remove firefox and decompress the cpio with busybox's cpio with -id i don't have firefox back.
If i do the same with geany package it restores smoothly.
So this should prove that there is something wrong with firefox package.
Maybe it has been packaged with the real cpio and this cause problems? Sure i can't umnderstand what's going wrong with firefox package.

ok, im (a little bit?) dumb. i did not mention that i'm using a liveusb nor i knew that using a liveusb the rootfs is extrated in ram by the kernel itself, so busybox and cpio are not involved at all.
reading this: http://www.mjmwired.net/kernel/Documentation/filesystems/ramfs-rootfs-initramfs.txt my attention has been attracted from this paragraph:

Note: The cpio man page contains some bad advice that will break your initramfs
archive if you follow it. It says "A typical way to generate the list
of filenames is with the find command; you should give find the -depth option
to minimize problems with permissions on directories that are unwritable or not
searchable." Don't do this when creating initramfs.cpio.gz images, it won't
work. The Linux kernel cpio extractor won't create files in a directory that
doesn't exist, so the directory entries must go before the files that go in
those directories.

This could prove that i'm sure not a linux guru but maybe it could be that something wrong happened when firefox have been packaged, or somewhere else. BTW wikipedia page about cpio reports some example with the -depth option.
Anyway, because i want to save my changes with tazusb and restoring without fear to screw up the system, i'm using this /etc/init.d/local.sh:

#!/bin/sh
# /etc/init.d/local.sh - Local startup commands.
#
# All commands here will be executed at boot time.
#
. /etc/init.d/rc.functions

echo "Starting local startup commands... "

#!/bin/sh
#are we root?
#if test $(id -u) != 0 ; then
#   echo -e "\nThis program requires being run as root.\n"
#   exit 0
#fi
#i check for the standard location of rootfs.gz or
#i mount my usbkey by label - put your label instead of mine if you
#have unusual setup

if [ ! -e /rootfs ]; then
	LABEL=CH32
	USER=`ls /home/`
		if [ -e /home/$USER/boot/rootfs.gz ]; then
			MOUNT_POINT=/home/$USER
		else
			MOUNT_POINT=/media/$LABEL
			if [ ! -e $MOUNT_POINT ]; then mkdir $MOUNT_POINT; fi
			if [ ! -e $MOUNT_POINT/boot/rootfs.gz ]; then mount /dev/disk/by-label/$LABEL $MOUNT_POINT; fi
		fi
	#unzlma rootfs.gz
	cp $MOUNT_POINT/boot/rootfs.gz / 
	mv /rootfs.gz /rootfs.lzma
	unlzma /rootfs.lzma
fi
#remove previous missing.txt
if [ -e /missing.txt ]; then
	rm /missing.txt
fi
cd /
(
IFS=$'\n'
cpio -t < rootfs | grep -v /dev | sed '$d' > list.txt
for i in `cat list.txt`
do
	if [ ! -e $i ]; then
		echo $i >> missing.txt
	fi
done
#restore missings from missing.txt
for i in `cat missing.txt`
do
	if [ ! -e $i ]; then
		cpio -i < rootfs $i
	fi
done
)

rm /rootfs	
rm /list.txt

not so elegant but it works.
this is my last missing.txt, if someone would like to check:
http://rapidshare.com/files/365004311/missing.txt.html

EDIT: i don't know how to post local.sh in a readable way, you can find it here
http://rapidshare.com/files/365006514/local.sh.html

It turns out that the problem is that i have two neurons which conflicts each other :-(
I was booting with the sound=no option to save some ram, so /etc/init.d/hwconf.sh remove sound modules and sound applications.
firefofox package depends on alsa-lib so it got removed with alsa-lib
You can verify that this time i'm right booting a slitaz cooking firefox flavor with the boot option sound=no
Proposed solutions:
1) remove alsa-lib from firefox dependencies (i'm signaling it in http://labs.slitaz.org/projects/packages)
2) modify /etc/init.d/hwconf.sh to remove sound packages but not packages depending on it. this would avoid futures incompatibility too.