When your browser starts up and you request a kernel, make sure that you specify the option
-mt x
in the args field, where 'x' is the number of processors available. There must be a space between the '-mt' and the 'x'.
If you don't get too much parallelism out of your tree too many calls to aips++, (from UVWs, coordinate conversions and such), may be the bottleneck -- this should become apparent from looking at the profiler. Also, I recommend installing a little utility called "gkrellm" to keep track of your CPU loads.
See also How do I get a node to poll its children in parallel?