User Agent: Mozilla/5.0 (X11; Ubuntu; Linux x86_64; rv:21.0) Gecko/20100101 Firefox/21.0 (Beta/Release) Build ID: 20130512194354 Steps to reproduce: I Have a parallax background made of generated fragments with random positions/opacities/colors AND sizes. (~20/30 fragments) (linux3.5 X86_64) Actual results: everything is okay, everything work BUT : - If I use "background-size: 100%" and/or "transform: scale(xx)" to randomize the size of the fragments, my CPU usage is 100% all the time ! Expected results: If I only use the "width+height" instead of "transform: scale()" properties the same result happens and the cpu Usage is this time nearly null... The problem is that I can't use the advantage of "background-size:100%" for the fragments which must fit in the size of it's area =/ (and I must then use a background-image trick =/) (I made the test under chromium and the cpu usage was very low in both cases)

Provide a testcase, please.

Here's two links, the properties are applied from the ".tesselle_1" to the "tesselle_11" elements. with background-size: 100%; and transform: scale(); : http://latortue.iriscouch.com/testcase_bug/_design/moreje/_rewrite and without those properties : http://latortue.iriscouch.com/testcase_bug/_design/moreje/_rewrite (if the page doesn't load, refresh multiple times to stress the server, it will respond ^^) I hope I did my report right, i'm not used to do that ^^"

oups, the second link is this one : http://latortue.iriscouch.com/testcase_nobug/_design/moreje/_rewrite

Yes, same important CPU use with FF21 on Win 7 (30%). Disabling HWA doesn't change anything.

Seems like same issue as https://bugzilla.mozilla.org/show_bug.cgi?id=846315 The analysis has evolved better in 846315, it's already identified the exact regression window.