We introduce a type system for concurrent programs described as a parallel imperative language using while-loops and fork/wait instructions, in which processes do not share a global memory, in order to analyze computational complexity. The type system provides an analysis of the data-flow based both on a data ramification principle related to tiering discipline and on secure typed languages. The main result states that well-typed processes characterize exactly the set of functions computable in polynomial space under termination, confluence and lock-freedom assumptions. More precisely, each process computes in polynomial time so that the evaluation of a process may be performed in polynomial time on a parallel model of computation. Type inference of the presented analysis is decidable in linear time provided that basic operator semantics is known.