I have seen more than 30% performance regression for LongHTML sub-test within
Swingmark benchmark for jdk1.4 b72 (comapred to jdk1.3.0). This regression started to appear in b71.

To reproduce:

% uname -a
SunOS vispro 5.7 Generic_106541-16 sun4u sparc SUNW,Ultra-30

% jdk1.4/bin/java -server -version
Java(TM) 2 Runtime Environment, Standard Edition (build 1.4.0-beta_refresh-b72)
Java HotSpot(TM) Server VM (build 1.4.0-beta_refresh-b72, mixed mode)


% jdk1.4/bin/java -server -ms256m -mx256m -cp /home/zzhang/bugs perfteam.swingmark.SwingMark -r 5 -q -testfile HTMLTests.txt

ShortHTMLTest 46730 21508 19081 19121 18843
ShortTextTest 20920 17774 16481 16391 16349
LongHTMLTest 336758 316112 315432 315466 313311
LongTextTest 115982 114531 113961 113783 113509

% jdk1.3/bin/java -server -version
java version "1.3.0"
Java(TM) 2 Runtime Environment, Standard Edition (build 1.3.0)
Java HotSpot(TM) Server VM (build 1.3.0, mixed mode)

% jdk1.3/bin/java -server -ms256m -mx256m -cp /home/zzhang/bugs perfteam.swingmark.SwingMark -r 5 -q -testfile HTMLTests.txt

ShortHTMLTest 55941 31280 31211 30890 30895
ShortTextTest 23673 19663 18894 19483 19093
LongHTMLTest 250160 226959 227581 226999 226107
LongTextTest 136250 133493 132603 133094 132233

You would see the regression in LongHTMLTest only.

replace -server with -client would get the numbers for client side.