Re-reading the original post I have an idea what might be the problem.
Someone has a fire engineer who has created a mythical 5s detection time, and it seems that this is being used to support a particularly low (read: unlikely) claimed evacaution time of <121 seconds. If elsewhere in the very same report the smoke temp is shown to be only just getting to 30C at the time the report is claiming evacuation should be complete, then it doesn't take a genius to suggest that the 5S detection time might be a little bit of an exaggeration.
For info: To work out a smoke detector's activation time the detector is often assumed to be a heat detector with a low RTi (5), and an activation temp of 10C above ambient. Based on this, as far as any calcs would go, they are quite right in suggesting that while the temp is lower than this 31C, detection may be delayed, but this is (as Stu points out) normal behaviour of any growing fire, not true stratification. They are simply using this to show that the 5S is improbable, and using a comment from BS5839 to back that up, not that the fire alarm system is wrong in any way.
IMO the approval body should be going straight for the details behind the ASET time calculated by the engineers. The time to detection, if calculated properly, will be way above 5s. After that they should consider the time that it will take to raise the alarm (typically instantaneous if no investigation time is factored in to the alarm) the time it takes people to react, (anything from 30s upwards depending on the occupancy) and then the walking time to the exits, and/or any queueing behaviour expected.
I think that this is strictly the fire engineers problem (maybe a problem on paper only) and is not a shorftfall in the alarm provision. If for any reason you need detection quicker than the 120S (approx) in order to evacuate people particularly quick, then the risk level seems to be beyond the scope of normal detection requirements and something such as aspirating detection or a way of controlling the risk more effectively is required.