[PATCH] benchtests: Cleanup bench-malloc-thread

[Wilco Dijkstra]

Hi DJ,

> Does changing the duration invalidate any existing results, or are
> benchtests always a "build twice and compare" system?
>
> I'd be interested in seeing the standard deviation for 10s vs 3s
> results.  Malloc times tend to be more chaotic than other parts of libc.

STDEV is slightly lower for the 3s variant, but it's very low anyway. 
For 10 runs on 1 thread it's less than 0.1%, for 32 threads it's less than 0.2%.
Overall scores are within 0.1%. The 3s variant still runs 0.5 billion iterations
which is already way overkill and many times more than the number of
iterations we use for string tests.

Cheers,
Wilco