I have to ask: Why is it so important to people that the libthr performance gains be impossible to achieve without use of the 1:1 model, rather than a modification of libc_r, or avoidance of existing kernel latencies? -- Terry