My guess is that you actually have generalized I/O contention and/or a higher-than-you-think paging load. Although you say that you have seven processes running, it seems to me that this workload is purely
That would account for the ~40% idle time but it also might account for other delays too. If the work that is being done basically does not involve the CPU, it can be counter-productive to parallelize it. I really don't expect that IPC would be the actual explanation.