>For the blissfully ignorant, what real world performance difference is
>this going to make?  Will TCP stack be "much faster"?

Sorry, I only have LMbench [modified, the %!# icache-stomping code
causes gas to punt] numbers for both this change and the faster
le_ioasic copy code together.

You could look at Jon Kay's Usenix 93 paper; so far it's basically the
same code.