Consider the following source code (BeanA and BeanB source is the same):

@SessionScoped
public class BeanA implements Serializable
{
@PostConstruct
public void init(){
System.out.println("A Created ");
}

public List<String> getA(){
return Arrays.asList("A", "A");
}
}

and the view:

<ice:panelTabSet>
<ice:panelTab label="1">
</ice:panelTab>
<ice:panelTab label="2">
<ice:dataTable value="#{beanB.b}" var="_b">
<ice:column>
<ice:outputText value="#{_b}" />
</ice:column>
</ice:dataTable>
</ice:panelTab>
<ice:panelTab label="3">
<ice:dataTable value="#{beanA.a}" var="_a">
<ice:column>
<ice:outputText value="#{_a}" />
</ice:column>
</ice:dataTable>
</ice:panelTab>
</ice:panelTabSet>

If you navigate to tab 2, the first bean´s init method will be called ("B Created"). Then navigate to tab 3 and you will see the same for A ("A created"). If you swap out the h:dataTable/h:column and replace it with it´s ice equivalents and retry the above test, both B and A are created at the same time when you click on tab 2 (("B Created" & "A Created"). Having all getters called can hurt performance since they may contain expensive logic such as DB accesses, etc..