Run the following test on Windows platform using both debug and release Tiger or Mustang builds (the same test is attached to the bug report):

---- BEGIN OF SOURCE CODE ----
import java.applet.*;
import java.awt.*;
import java.awt.event.*;

public class StressNotify extends Applet implements ActionListener {

Button onceButton, manyButton;
LabelPanel labelPanelGreen, labelPanelRed;
int panelShowing = -1;
static final int RED = 0;
static final int GREEN = 1;
Panel panel;

public void init()
{
onceButton = new Button("Change panel once");
add(onceButton);
onceButton.addActionListener(this);
manyButton = new Button("Change panel many times");
add(manyButton);
manyButton.addActionListener(this);
panel = new Panel();
add(panel);
addLabelPanel(GREEN);
}

void addLabelPanel(int panelToAdd)
{
panel.removeAll();
if (panelToAdd == GREEN)
{
labelPanelGreen = new LabelPanel(Color.green);
panel.add(labelPanelGreen);
}
else if (panelToAdd == RED)
{
labelPanelRed = new LabelPanel(Color.red);
panel.add(labelPanelRed);
}
panelShowing = panelToAdd;
}

public void actionPerformed(ActionEvent ae)
{
if (ae.getSource() == onceButton)
{
addLabelPanel(1 - panelShowing);
// panel.removeAll();
validate();
}
else if (ae.getSource() == manyButton)
{
// while (true)
                for (int i = 0; i < 10; i++)
{
addLabelPanel(1 - panelShowing);
validate();
}
}
}

public static void main(String[] args)
{
  StressNotify s = new StressNotify();
  Frame f = new Frame("F");
  f.setBounds(100, 100, 640, 320);
  f.add(s);
  f.setVisible(true);
  s.init();
  s.start();
}
} // END OF Class StressNotify


class LabelPanel extends Panel {

int labels = 100;

LabelPanel(Color color)
{
setLayout(new GridLayout(labels/20, 20, 4, 4));
setBackground(color);
Button[] myLabel = new Button[labels];
for (int i=0; i<labels; i++)
{
myLabel[i] = new Button(String.valueOf(i));
myLabel[i].setBackground(color);
add(myLabel[i]);
}
}
} // END OF Class LabelPanel

---- END OF SOURCE CODE ----

Open Task Manager and found the Java process. Press the button 'Change panel many times' several times and look at the process' memory usage. When running with release build the amount of memory used by Java process will grow for some time and then remain at the same level (about 26Mb on my workstation). However if the test is run with debug build, memory usage will grow until all the system memory is allocated and then the test hangs.

This is not a recent (Tiger or Mustang) regression, the problem is reproducible at least with 5.0u5, 6.0b61, 6.0beta and 6.0beta2.

Also, I've noticed that the flag -Xmx doesn't affect this behaviour, so it looks like some memory leak in the native code. I'm not sure if this an AWT or VM issue, it needs to be investigated.