Non-determinism of the test was caused by lack of setting a proper value to hb period,
actually fixed by BUG@50767.

These fixes aim at possible non-determinism in comparison of received
hb events by master and slave in the circular part of the test.
Even though the HB periods ratio was choosen to be as high as 10, it's still incorrect 
to compare number of hb-events basing only a relation between their periods.

Yet another issue is relatively short 60 secs timeout of wait_for_status_var.inc
makes valgrind runs to fail. 
Fixed with deploying wait_for_slave_io_to_start afront of calling wait_for_status_var.

The test is made runnable only with MIXED binlog-format as it has close to 1 min
total exec time and there is nothing format specific in it.