Thanks Sean for the suggestions !

Indeed it works, but it is still not 100% of multi-threaded work.
At least in my experiments, I run an ordinary non-master slave
multi-threaded problem on a single machine, and
compared it with the same multi-threaded problem run in master -
multi-threaded slave.
Non-master slave multi-threaded  problem on a single machine used the
multi-threads at their full-capacity.
Looks like because of such workarounds, the multi-threaded  benefit in
master - multi-threaded  slave could not be used in its full capacity.
At some point of time all the threads are running, at some point only single
thread is running by the slave. I think that 
breeding at least and some other initializations by the problem are done in
multi-threaded  way, however I am not
sure that this can be said for all of the phases by evaluation, pre- or post
evaluation.
At least this is my overview. It could be also wrong. But I hope that the
master-multi-threaded slave feature will be soon added to the ecj, so the
whole capacity of multi-core machines can be used by master-slave
configurations.

Thanks once again,
Nikola