Probably because I moved CSSPropertyDisplay from the fast table to the slow table.
http://webkit-perf.appspot.com/graph.html#tests=[[39020,2001,3001]]&sel=1337320768437.756,1337421568437.756,3097.560975609756,5841.463414634147&displayrange=7&datatype=running
Created attachment 143334 [details] Patch
This seemed to fix the regression on my machine, but I see a 5% swing in the test results locally, so it's hard to be certain (the graph shows similar variance).
Comment on attachment 143334 [details] Patch Clearing flags on attachment: 143334 Committed r118044: <http://trac.webkit.org/changeset/118044>
All reviewed patches have been landed. Closing bug.
It does appear that this patch fixed the regression: http://webkit-perf.appspot.com/graph.html#tests=[[39020,2001,3001]]&sel=1337739099058.127,1337757519362.6953,3278.688524590164,6426.229508196721&displayrange=7&datatype=running
This caused bug 105275.