I haven't used Sonuus, but recent observations suggest that the capability to render multi-part audio to notation may be very near on the horizon.
Melodyne (the pitch-correction people) are putting out a version that purports to identify all the notes in a polyphonic audio clip and correct them independently. One would think that the ability to computer-ly recognize the notes would lead directly to the ability to put them out as notation or MIDI.
As far as the horror raw MIDI recordings make of the notation, leeneia, I've found in my messing about with sequencers that the right "quantize" settings can handle 90% of the cleaning-up. The last bit will always be up to the human, but it can get rid of most of the mess and give you something to work with.