The fix for RT-19532 "TitledPane arrow is shrinking when styling" which has been integrated into
fx2.1-controls-scrum-b345 lead to following performance regression:

 Controls.Buttons-adhoc-mouse-nodes1008-skipAll: 167.98 ± 0.04 -65% (-317.01)

Smaller regressions are:
 Controls.CheckBox-adhoc-items300-toggleAll: 23.57 ± 0.02 -11% (-3.01)
 Controls.RadioButton-adhoc-items300-toggleAll: 36.52 ± 0.16 -16% (-6.86)

RT-19532 changes are:
> diff -r a5ca94ced4cd -r 5a3062b456a5 javafx-ui-common/src/com/sun/javafx/css/StyleHelper.java
--- a/javafx-ui-common/src/com/sun/javafx/css/StyleHelper.java Wed Feb 08 11:21:23 2012 +0100
+++ b/javafx-ui-common/src/com/sun/javafx/css/StyleHelper.java Thu Feb 09 17:41:29 2012 -0800
@@ -1023,7 +1023,7 @@
                 final ParsedValue parsedValue = style.getParsedValue();
                 if (!parsedValue.isLookup() && !parsedValue.isContainsLookups() &&
                     // RT-19192 - if -fx-font-size is relative, do not cache
- !(resolved.isNeedsFont() && property.endsWith("font-size"))) {
+ !(resolved.isNeedsFont())) {
                     style.value = val;
                 }
                 

So, we do "style.value = val;" much more frequently now.

Would be nice to understand whether this is really proper fix.

Steps to run Controls.Buttons benchmark
> cd FX_WS/tests/performance/FXBenchmark
  ant
> cd cd FX_WS/tests/performance/MouseResponseTest
> ant
> java -Djavafx.animation.fullspeed=true
  -cp "JAVA_FX/rt/lib/jfxrt.jar;../../../import/benchmarks-2.1.1/benchmarks-2.1.1.jar;../FXBenchmark/dist/FXBenchmark.jar;dist/MouseResponseTest.jar"
  jrockit.bm.Main uiresponse.bm.MouseResponseFpsBenchmark
  -i 1 -wt 0 -tr 30 -nodes 1008 -skip 1007 -click-mode first-last