If your cluster is set up primarily as a job farm then SGE will have been set up with 2 job slots per compute node. In this case you need to submit a multi-threaded job to a special parallel queue to ensure that it occupies 2 slots on a single compute node rather than 2 slots from 2 compute nodes. e.g.
qsub -pe smp 2 my_2thread_job.shwould submit a job to occupy both processors on a dual proceessor compute node.
In this case if you were to submit a multi-threaded job as a scalar job then it is possible your code would share with another scalar job.
Please ask your administrator which way you should submit a multi-threaded job if you are unsure how your cluster is set up.