1½ Topics: realtime throttling and user-space adaptive spinning

I'm not a realtime expert, but my first impression is that if your realtime tasks are getting 95% of the CPU and you're still missing realtime deadlines, you've overcommitted the CPU and need to re-design your app. That workload is, IMHO, running without sufficient safety margins for a true hard realtime system.

However, it appears that the the article is describing a soft realtime system (a browser?), and they're trying to eke performance out of it. That seems contradictory to the goals of realtime. That is, usually there's tradeoff between realtime and throughput, where you only get to prioritize one or the other. Am I missing something? (I'm sure I am.)