benchmarking confusion

[Stephen Weeks]

I am confused.  I decided to look into the vector-rev benchmark that had the
worst ratio (0.8) from last night's benchmarks.  In looking at the code, I
noticed that the assembly code for the hot loops is identical with and without
local flattening.  Furthermore, I ran the benchmarks on my machine and noticed
that local flattening gives a noticeable speedup, not slowdown.

		eponym	starlinux
local-flatten 	time	time
true		7.3	3.7
false		7.8	3.5

How can the assembly code be giving such differences on one machine, and, why
might I see such variance across two machines.