oprofile shows the test URL spends about 30% of the time in libgtklayout.so. A debug build prints the following message at what looks to be roughly one per second: nsBlockReflowContext: Area(div)(1)@0x8329c64 didn't set max-element-size!

The patch in bug 192267 makes the message go away, though it still uses about 30% of the cpu in layout.

Basic breakdown of time on that testcase: Total jprof hits (non-realtime): 2305 Reflow (under PresShell::ProcessReflowCommands): 275 Painting (under nsViewManager::Refresh): 606 of which under nsImageFrame::Paint: 548 and the rest various view manager and frame overhead In detail, under nsImageFrame::Paint we have: 210 hits under RectStretch 204 hits under nsImageGTK::DrawComposited32 the rest mostly stuff under nsImageGTK::DrawComposited but outside RectStretch and DrawComposited32 Setting of style.left and style.top: 768 of which: 117 is in the CSS parser 391 is in style reresolution 161 is in ProcessRestyledFrames That's the big stuff....

This sounds somewhat related to bug 157681

Is this worth reprofiling ?

test page down so it will be very complicated to reproduce this can be closed i think