I don't know if it's possible, but I would like to see different polyphony on different "modules" within an arranger or a workstation. I think the Korg Kronos does this but with built-in soft synths running in Linux.

Within one board, you could have these...
128 Note Piano Module
64 Note Organ Module
128 Note String Module (Different Articulations)
64 Solo Instruments (Like the SA2 and DNC)
128 Note Synth Module (Other Sounds Needed)
64 Drum/Percussion Module

These numbers can be any thing that works. I have full orchestrations that sound very big coming from 32 note polyphony. I can be done but the multi layering is the problem. A string sound sounds very thick... but you are using 4 layered multi samples to make the thick sound. Put a few more fingers down and a sustain pedal and bye-bye first note!!!

Each one working on it's own polyphony system. They can even call it Multi Polyphony Systems... MPS!!

Memory is always getting cheaper! Be the brave company to break the "glass ceiling" of polyphony and give us power for our $4000-$5500 spent! Then spend your time supporting and upgrading the unit instead of replacing it every 2-3 years.

Be Brave and take the Leap!!!

All SC sound engine based Roland keyboards handle polyphonic well..128 polyhony is fine, as most of the early 64 were..

Per arranger the polyphony is handled in this way..Priority goes to drums first (always allowing a mininum of voices), than bass, etc for the most important parts..each with a mininum voice..

The next important way to handle polyphony is last note reserve ..that allows the first note to drop off, making it almost impossible to hear..

I have never heard a drop off on any 128 voice Roland arranger that I have played..

I have had drop offs with Yamaha arrangers (low priced PSR's would mute all sounds till it recovered..what a nightmare).yet a Roland 24 and 28 voice SC board (E-66), would have no noticble drops, thanks to priority voice allocation..

Todays 128 Roland's are more than enough in the arranger line..

I should mention the MediaStation was 256 voice, never dropped out, even though it didn't seem to have the advanced allocation Roland uses...it relied on pure numbers..

G1000 64 voice was fine, as the Korg 62 voice...I recall the 126 voice Yamaha 9000 Pro , did not fare well in layerd pianp/strings...very noticable.(probably poor allocation method)

Another consideration that will save polyphony..Use of sample "parts" (Roland).."elements" (Korg), and whatever Yamaha calls it sound events..

Most of your better sound designers use 1 or 2 sample parts (elements)..keep in mind it will use a polyphony part for each element triggered..The use of velocity switched sounds and velocity filtering save polyphony..Better sound design takes advantage of this..

The key to using less parts/elements are bigger , fuller samples..limiting the need for more partials (elements) needed..
Filtering and changing of the larger samples does not use more polyphony, as added parts and elements do..

Here again is where the Roland G70 or BK series shines, most are 1 or 2 partials (complete samples)..compared to as an example 4 element Korg sounds (use 4 polyphony) for a single sound..
Velocity switching sample tones is a big reason Roland handles polyphony better than it's competition..

Just some things to think about, while you try to understand the polyphony issues..