They are used in entirely different scenarios.
Consider this scenario:
A Purchase Order has come from a Business Partner to your company and you start an orchestration to send out an e-mail to corresponding department. However, you can not consider the business cycle is complete until ordered item is dispatched. May be your orchestration will want to wait until a sales agent from your company has confirmed the dispatch via another message. This time gap can be 1 hour or even 1 day. Until then (time you started orchestration and the sales agent confirms it) you may want your orchestration to sleep and wake up again upon receipt of the second message.
In the above scenario, I'll use correlation set.
Consider this scenario:
Your company is a stock quote provider. Apart from real time data (Your Company has already built up a generic orchestration to fetch real time data from NASDAQ.), your company makes some value addition to the message by adding 52 week highs and lows of the stock. In this case, I'll build a new orchestration which calls the already built orchestration and once that returns the real time quote (the time gap is really very less here) new orchestration will continue to do value addition by hitting your company's database.
In the above scenario, call orchestration is a perfect choice.