People of course object that UNIX can do just as well. You can't prove it by the numbers.

This comparison was measured on a 120Mhz Pentium machine. Note that between 4k and 8k transfers you can see the effect of self-interference in the 8k data cache.

The figure compares local data transfer using a variety of IPC-like mechanisms. The Linux measurement was from the RedHat 4.2 release, which was a late 1.0 Linux kernel. Current pipe bandwidth figures are better, but still about 25% slower than EROS IPC bandwidth.