If you constrain 'performance' to things which can be easily measured and then test how multitasking affects those measurements, you can come to some broad conclusions that are also useless.