Spin off flows achieve greater throughput by allowing assignment to be worked on by multiple users by acquiring separate locks for each sub process.
Spin off flows are executed in the current requestor in a serial manner, however they do provide business parallelism the users to have assignments on the same work object.
I think the first statement is true, but on a website, it says the second statement is true.
Spin off flows do not create separate work items, there is still just the one work item so only one person at a time can hold a lock on it.
Spin off flows create multiple assignments on a work object and allow those assignments to be picked up in any order. This can improve throughput by allowing any person with free time to address their part of the work. With a single linear flow, if one person is busy or out-of-office, they would potentially block all of people with tasks setup later in the flow from getting their work done.
Spin off flows are useful when there are multiple tasks that do not depend upon the results of each other.