You are probably in the realm of asynchronous parallel computations where units of work are performed by processors at their pace and communication is performed explicitly. This model is a good approximation to many real life parallel computers such as PC clusters or multicore CPUs.
You have an algorithm that can be represented as $O(n \log n)$ units of work each taking constant time or as $O(n)$ units of work each taking $O(\log n)$ time.
Here $n$ is a parameter that characterizes the size of the problem.
The units of work can be executed on a parallel computer with a fixed number of sequential processing elements (e.g. processor cores). It depends on the algorithm whether the work units can complete while other work units have not started, or whether the computations will have to interleave.
In a practical computer interleaving can be achieved through pre-emption and context switches.