It can be explained by the fact that Opera isn't the main rivals of Firefox in the browser market.

Edit: http://arewefastyet.com/faq.html

  The team's explanation: Right now, the performance tests
  are run on a Mac, which means no IE. Also the tests rely on
  a "shell" JS engine that runs in a command line. It doesn't
  test browsers. We'll change that, eventually.

Can you provide any evidence to support this?

The only metric I've ever seen Opera beat Chrome on was IE9's flying images benchmark.

I do a lot of experiments around the edge of browser performance (and also regularly check other folks' demos in multiple browsers).

Benchmarks are not completely representative to "real world" performance. There are always some areas which work very well in particular browsers and areas in which some browsers are absolutely terrible.

You can get practically arbitrary order of browser performance by mixing "do-good" and "do-bad" tasks.

Subjectively, since recently Opera often feels faster than Chrome [1].

Not always, it's maybe 50/50 split, sometimes Chrome is faster, but in overall subjective experience Opera feels better (by a tiny bit).

I suspect it's because Chrome excels mainly at "behind-the-scene" more numerical type of tasks, while Opera is faster at more tangible "in-your-face" things like DOM manipulation / rendering / canvas.

Also, again subjectively, Opera's performance feels better "balanced". Chrome is crazy fast in some areas (see V8 benchmark), but slow/average in others. Opera is fast almost everywhere, so it has less bottlenecks.

-----

[1] More objectively, from my stuff which has numerically comparable performance:

http://alteredqualia.com/canvasmol/#DNA_crystal

On my notebook, this makes:

- 29 FPS in Chrome 7.0.517.0 dev

- 42 FPS in Opera 10.62 (45% faster)

Though for some simpler molecules, Chrome is sometimes much faster. Go figure.

-----

Another test:

http://alteredqualia.com/attractor/#TPsP2jWzkUx22cuNN5XToJNV...

- 525 ms Chrome

- 483 ms Opera (9% faster)

[1 math unit, 1 rendering unit]

-----

Case for "Chrome-has-faster-math", above example with "high detail" (which does more calculations compared to rendering):

- 1611 ms Chrome

- 2152 ms Opera (34% slower)

[10 math units, 1 rendering unit]

-----

Above with "wallpaper" settings (much more math, but also significant rendering):

- 4304 ms Chrome

- 5116 ms Opera (19% slower)

[20 math units, 4 rendering units]

It's ahead in the Peacekeeper benchmark, but falls just short on the Sunspider and JS tests.

http://www.geek.com/articles/news/opera-10-6-tops-performanc...

The Peacekeeper benchmark has some glaring flaws in the benchmark code:

http://weblogs.mozillazine.org/bz/archives/020664.html

(For example, "The benchmark starts with an array of 100,000 elements and removes 20 elements for each call. After 5000 calls, the calls become no-ops, and the benchmark then times those no-ops.")

Gah, I was about to say that Opera beats Chrome at Sunspider on my machine, but then realized I was still running Chrome 5. Just upgraded. Here are the Sunspider numbers on my machine:

Chrome 5: 315

Opera 10.60: 287.2

Chrome 6: 279.2

Point taken.