Oh, no the segfault happened when using qemu/serialice v1.5 and is solved using qemu from svn, on 32-bit hardware too.
In my previous message I forgot to attach the superiotool dump -- see attached file.

Following quote is after soft reset, typing some text and hitting the reset button:

SerialICE v1.5 (Aug 27 2010)

> 1
> 2
> 3
> 4
> 5
> 6
> 7
> 8
> 9
> 0
> r
> e
> s
> e
> t
> -
> -
> >
> òóñôõöôöòñðôõô÷ôôöðôÿôõõôõôô

Since the southbridge and superio datasheets mention the existence of two serial ports, I followed their guidance.
I thought that (the console printing part of) SerialICE, when setup the correct way, should survive a hard reset/power cycle regardless of the qemu part is running or not.

Since I don't have an oscilloscope, I've tried setting CLKSEL to 24 MHz and 48 MHz:
pnp_write_register(SUPERIO_
CONFIG_PORT, 0x24, 0xb4); // 24 MHz and KBC=1
pnp_write_register(SUPERIO_CONFIG_PORT, 0x24, 0xc4); // 48 MHz and KBC=1

What information is leading, the info from the superio or the info from the southbridge ?