Using r80210, the following demo takes 8619ms whereas it only takes 2505ms on Safari 5 http://people.mozilla.org/~roc/scrolling-boxes.html
r73316: ok r74228: ok r75294: bad r75891: bad r76640: bad
<rdar://problem/9119011>
It seems like there have been multiple regressions. Due to one of these two changes: http://trac.webkit.org/changeset/74538 http://trac.webkit.org/changeset/74539 I see a 140% slowdown (~2000ms to ~4800ms). There is another slowdown somewhere between r74539 and ToT, which gets times ~8000ms. I'll see if I can track down that regression as well.
The second regression is due to <http://trac.webkit.org/changeset/77101> I think.
The first regression is caused by <http://trac.webkit.org/changeset/74538>, which turned on support for non-prefixed box-shadow.
I filed <https://bugs.webkit.org/show_bug.cgi?id=56162> to track the second slowdown.
The demo uses non-prefixed box-shadow, so presumably if it had used -webkit-box-shadow instead, this demo would have been slower in Safari 5 as well. Since we just weren't applying the box-shadow before r74538, this doesn't seem like an actual regression. I think <https://bugs.webkit.org/show_bug.cgi?id=56162> is still valid though, especially since r77101 claims to improve scroll performance.
Yea, Safari 5 gets the same performance numbers as r74538 if I change the test to use -webkit-box-shadow instead of box-shadow.
I think that fact makes this bug invalid, but <https://bugs.webkit.org/show_bug.cgi?id=56162> still seems like a legitimate regression.
Thanks for the analysis, Andy!