flip-flopping between parallel and single-thread code

[pedr...@lycos.com]

in geophysical processing, you have both single-threaded and parallel-safe
modules. Data flow may go like this:

input-parallel-single-parallel-output

Now the single-threaded module may be defined so simply because it is so
trivial and the programmer thought the message passing overhead was not
worth it. But then does this not mean that the data is merged going into
it, and then sprayed out again to multiple cores?
So would it not be better to remain in parallel mode, by enabling the
single threader to run parallel?