Here's the (draft) - final report - which explains the rationale behind the conversation test they used.
They echo your point - "There may be some concern that the phone conversations in this study were harder than typical in-car conversations. The material was difficult and the conversations, although not continuous, lasted for the duration of the test drive."
Read the report for the follow-up 'however...'.
Regardless of whether you consider this a valid test, they fairly convincingly show that for certain mobile phone tasks, a test subject's driving performance - albeit in a simulator - is worse that when the same driver is over the legal alcohol limit (0.08% in the UK).