I have problems with the time measurement on  21264 Linux
systems.
I have seen output like:

[frohlich@spe85 frohlich]$ date ; time 1>/dev/null ./tstl95
; date
Fri Aug 27 04:49:35 EDT 1999
275.38user 0.00system 0:34.65elapsed 794%CPU
(0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (92major+372minor)pagefaults 0swaps
Fri Aug 27 04:50:10 EDT 1999


Where you can see that the elapsed CPU-time is about eight
times longer then
the real time! It seems to me that there is an
inconsistency between the
system headers which define the number of clock ticks to
1024 and the
return value of times (libc-system call) which seems to
return eight times
more ticks a second ...

I did some small timings with this machine and was very
surprised that
the linux system has about eight times smaller MFlop rates
(with the
same assembly code) then the True64 system!