The advantages of this include full test isolation and the fact that a test can now cause a PHP fatal error or even a segmentation fault of the PHP interpreter without interrupting the test execution.

He does mention, however, that it can cause a bit more overhead for larger testing suites since it needs to create a new process (complete with memory usage) for each running test. There are also coding issues that could be thrown off by this option (he gives an example of an inheritance issue with eZ Components).

He also notes some of his thoughts on how to control/configure the process (like in a configuration file) and via an "@isolated" notation to make it easy to do it dynamically.