Testing Two Mapping Libraries - Orika Vs JMapper

I have downloaded and testing these two mapping libraries. I wrote a program which has 100000 iterations and maps the beans of the same class:

public class IntBean {
 
    @JMap
    private int int1;
    @JMap
    private int int2;

[Code] ....
 
Mappers are created BEFORE iterations start:
 
    private JMapper jmapper = new JMapper(IntBean.class, IntBean.class);
    private MapperFactory orikaFactory = new DefaultMapperFactory.Builder().build();
    private MapperFacade orikaFacade = null;
orikaFactory.registerClassMap(orikaFactory.classMap(IntBean.class,IntBean.class).byDefault().toClassMap());    orikaFacade = orikaFactory.getMapperFacade(); 

What is in each iteration:this.orikaFacade.map(a1, a2);
or
a2 =  (A) this.jmapper2.getDestination(a1);
 
I know, that Orika and Jmapper are great libraries from Google and they use reflection in a different way than for example Dozer, which is much slower, they se reflection to generete code somehow..

I have 3 questions:

1) How they work - when the code is generated, during maven build, in runtime - everytime when I create mapper in code? Are they change class code byte dynamically?
2) Why there is this speed difference that I noticed?
3) Which library would you choose and why? Both have the same capabilities? Why both come from Google? Why Google didnt develop Orika and created Jmapper instead?