System.arraycopy()-related performance regression

(Note: below is original bug description; see evaluation section for
a partial diagnosis and simplified test cases)

I have seen more than 20% performance regressions for some serialization benchmark sub-tests:
   Byte arrays
   Int arrays
   Long arrays
   Float arrays
   Double arrays

(compared to jdk 1.3.0).

To reproduce:

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

% jdk1.4/bin/java -version
java version "1.4.0-beta_refresh"
Java(TM) 2 Runtime Environment, Standard Edition (build 1.4.0-beta_refresh-b70)
Java HotSpot(TM) Client VM (build 1.4.0-beta_refresh-b70, mixed mode)


% jdk1.4/bin/java -client -ms256m -mx256m -cp /home/zzhang/bugs/rmi-tests/serialbench.jar bench.serial.Main


Byte arrays 850.0
Int arrays 3350.0
Long arrays 8351.0
Float arrays 2930.0
Double arrays 5174.0

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

% jdk1.3/bin/java -client -ms256m -mx256m -cp /home/zzhang/bugs/rmi-tests/serialbench.jar bench.serial.Main

Byte arrays 552.0
Int arrays 2563.0
Long arrays 5401.0
Float arrays 2163.0
Double arrays 3311.0