With a defined artificial latency headroom and a message parameter to (partially) override that latency it would be possible. With reasonable low maximum concurrency and a very high bandwidth/message size ratio the required extra latency could stay well within the range of centimeters at speed of sound (a very high bandwidth/message size ratio would certainly also lessen the cost of not taking that feature).

The next roadblock on the way to truly concurrent chords would probably be controller readout. I know nothing about how those are typically implemented, but I strongly suspect sequential readout.