TBB pipeline FAQ says that multiple TBB pipelines can be run concurrently if they are launched from different TBB tasks or OS threads. I would like to clarify the meaning of OS threads: do you mean logical threads, as can be created by pthreads, say?
In general it is safe (from both correctness and performance standpoint) to run TBB's parallel algorithms from wherever you are, independently on whether other TBB algorithms has already been started from other threads. It is also safe to run one algorithm inside the other (nested parallelism). TBB efficiently handles all these usage models.