the latency is about 7~8 us roundtrip between ppu -- spu
There are people getting 0.3~0.5 us reported in 
http://www-128.ibm.com/developerworks/forums/dw_thread.jsp?message=13933627&cat=46&thread=156160&treeDisplayType=threadmode1&forum=739#13933627
I tried it and verified that.
12/03/2008: the code is updated to run with libspe2, sdk3.0.
the performance now is 0.27 us roundup trip using problem state  to send and receive messages