Gelato@UNSW is part of the Gelato federation's Scalability focus group.
The focus group's aim is to improve Linux's ability to scale from the tiniest PDA to the largest SMP or ccNUMA machine available.
There are some nice benchmark figures at http://bulk.fefe.de/scalability/ showing how Linux and other free operating systems scale for webserving. Linux 2.6 comes out quite well, as many of the algorithms that are important are O(1).
The Linux Scalability Effort (LSE) group is also doing good stuff.
OSDL has a Scalable Test Platform project as well.
It doesn't appear that there are any really good scalable benchmarks that everyone agrees are the best. Reported figures of merit seem to be from taking the core of some application or other and running it on varying numbers of processors, amounts of memory, etc., to see how it (and the underlying machine and operating system) scale. Most synthetic benchmarks do not scale themselves, so do not test the scalability aspects of the machine and operating system.
Gelato@UNSW has begun a project that will identify and rectify Benchmarks suitable for distributed systems.