Race with softlockup detector

As an SRE, I would be very unhappy with a solution along those lines. If you want to put something like this into production, I would expect it to have quite low overhead, because hardware and electricity are expensive and carry significant opportunity costs. And if you have low overhead, then the kernel's softlockup detection should not have a problem with you in the first place.

I wouldn't necessarily object to doing this sort of thing in a non-prod CI or testing setup, because that's (usually) much smaller and O(1) in terms of incoming QPS (i.e. if you have twice as many incoming requests, you do not need to turn up twice as many test machines, but you do need twice as many prod machines to serve those requests).