glibc-2.3.6-r3 had picked up the amd64 string and math patches as described in bug #100289. glibc-2.4 now at least lacks the string optimizations, I did not test math optimizations due to other problems which forced me to go back to 2.3.6-r3 (see bug #125909)
*** This bug has been marked as a duplicate of 100289 ***
Well, I almost expected this to become a duplicated, but decided to post a new bug (with reference to the old) anyways, since in my eyes this is something different. This time it clearly is a regression, as it was in portage before, admittedly depending on the same patches as before. That's why I consider this a different bug. If 100289 is finally handled properly and the patches there are reintegrated, this bug of course can be considered void.