Wavetrac Differential

[JT A.]

They have already tried most configurations for arbs, including stiff arb front and no arb rear.

This is what sticks out to me. If they have tried big changes to front & rear roll stiffness (I don't know what values you consider "stiff" and "soft") but the chassis doesn't respond to it, it's an indicator that your chassis' torsional stiffness is too low. Could be a global stiffness issue or it could be a local stiffness problem right at the bellcranks/bellcrank posts. You said 1800 ft*lb/deg was your torsional stiffness from FEA. If that includes the whole chassis & suspension assembly (hub-to-hub stiffness in other words) then it is probably acceptable, but if that's "frame only" then I'd consider it a bit more soft than desirable. If your analysis was over-constrained in any way to inflate your results, or any errors in welding/assembly could mean you're even further away from a good target.

I'd recommend doing a physical test of your hub-to-hub torsional stiffness. Or an alternative that may be easier to set up, you can measure your wheel rates on scales - if they're significantly softer than what your wheel rate calculations would predict, you've most likely got problems with installation stiffness of your belcranks.