Check it out:
http://lkml.org/archive/2004/10/9/12/index.html
MontaVista just posted their RT enhancements to the kernel. They
incorporate some of the VP patches (irq threading) but go beyond that in
several areas. They claim performance similar to the VP patches right
now, but whereas the VP approach has reached the lower limit of latency
(hundreds of usecs), due to replacing spinlocks with mutexes their
approach eventually should be much better (tens of usecs, bounded by
worst case IRQ disable).
Have not tried it yet but it looks very cool!
Lee