Comparation with PIL thumbnail() method

[davide-romanini]

I made some rough test on my NAS, an ARMv5TE based device without neither FPU nor NEON extensions, especially slow on media processing.
I'm trying to optimize thumbnail generation, but I wonder why benchmarks are comparing jpegtran to the resize method of PIL, that's really slow, instead of comparing with the thumbnail method of PIL, that's a lot faster.
In my brief testcase I found a smaller difference in performance: jpegtran seem around 20-30% faster but with worse quality, at least at first glance..
So I wonder if you made some benchmarking involving also PIL thumbnail.