Hmm... I have read http://www.archlinux.org/pipermail/arch/2006-July/011385.html but... there are:

umount /proc/bus/usb
umount /proc && mount -t proc none /proc
[ "`grep sysfs /proc/filesystems`" ] && umount /sys && mount -t sysfs none /sys
[ "`grep usbfs /proc/filesystems`" ] && mount -t usbfs none /proc/bus/usb
# now mount all the local filesystems
/bin/mount -a -t $NETFS

Which means in current initscripts usbfs is mounted ignoring fstab.
So, when you remove nousbfs from $NETFS then /bin/mount -a -t $NETFS will try to mount usbfs again?
Or did I miss something?

As Thomas Bächler said, specific mounting of usbfs in rc.sysinit should be removed, as well as nousbfs from $NETFS.
But then, -t nousbfs should be added after -t $NETFS in fsck step.
I'll post a patch for all this, then please tell me if it fixes your problem.

But... then /proc/bus/usb will not be mounted unless it's written in fstab! Can this hurt users with USB devices???

you can check for the existence an usbfs line in fstab, if it exists,
then your script must be aware of all the options passed in fstab, otherwise
it can mount usbfs with the standard options.

And how the script should check for the existence an usbfs line in fstab? And it must also differentiate commented and uncommented line. No, it's too complex.

Maybe as a workaround for this usbfs problem, you can just remount usbfs in rc.local?

yes, could be a solution even not so elegant ;-)
Thanks !

So this can be closed? Because it is hard to not break something while fixing this, IMHO.

ok, close it thanks :)