Issue has been fixed: Fix wrong rate limit + add a few logs. by eskimor · Pull Request #6440 · paritytech/polkadot · GitHub
The fix has been verified on Versi. More than 30 disputes per second are raised on average, network is still chugging along nicely. (Slower of course, but no other issues.)
The issue is described in the above referenced PR, in a nutshell it was a detrimental feedback loop. With the fix, dispute-coordinator is handling that large amount of disputes with just 20% CPU usage, dispute-distribution even only at 2%. The system is bottle-necking on candidate validation, which is how it is meant to be: Everything just gets slower with more load, nothing else happens.