The author of the POSIX.4 scheduler mods is Markus Kuhn; the archives or dejanews will have the announcements and performance utilities. I assume everything made it into the 2.0 kernel.
I did have occasion to compare the dispatch latency with real (microsecond-resolution) hardware timers. Once it's running under SCHED_FIFO and everything is locked down, latency is quite stable, though there were a few spikes up to a few milliseconds. I think this might have been some network code.
In general I don't think there's heavy emphasis on the part of kernel-driver authors to be careful about disabling interrupts for a long time. Your mileage will depend on what mix of kernel code gets run. Some sort of monitoring is a very good idea; I'm told the Pentium has a cycle counter on-chip, which is ideal.
Cheers, Peter Monta