The problem is if you apply standard BRM rules, you have to average 22kmh between dreux and ramboulliet if you leave dreux on the time limit...
The loading protects you from that rude awakening
The loading itself doesn't protect you from that.
They are two different things.
First of all, as you say, the ride is over distance, so if you do 13.3kph for 90 hours you end up some 25km short of the finish of a 1225km event. SR Steve's suggestion is to just use ~13.6kph (actual ride distance / 90h) as the minimum average speed for the whole event so there is no front loading required and the minimum average speed is flat over the entire ride and the last control times aren't shortened. But ACP/BRM rides do not do this.
The completely separate thing is why ACP/BRM rides "front load" rides with increased average speeds in the earlier stages and then lower average speeds later on.
Article 10 here goes into the basic BRM rules for closing times:
http://www.audax-club-parisien.com/EN/312%20-%20Rules%20of%20BRM%20France.htmlAdditionally, riders must arrive at each checkpoint between the opening and closing time for the checkpoint and calculated as follows:
Opening: 34 km / h (km 1 to 200); 32 km / h (km 201 to 400); 30 km / h (km 401 to 600); 28 km / h (km 601 to 1000); commercial rounded by the minute.
Closing: 1 hour + 20 km / h (km 1 to 60); 15 km / h (km 61 to 600); 11.428 km / h (km 601 to 1000); commercial rounded by the minute.
Of course, this still doesn't cover everything exactly as it would put the closing time for a control at 400km at 26h40 despite the official time limit for a 400 being set at 27h (14.815kph). Audax is more a list of exceptions to rules than rules themselves.
These figures also don't match the control closing times in PBP, I've no idea exactly what formula they use there but my best guess is in the thread I linked earlier.
The question is why BRM/ACP front load the controls so much, and my guess is that it's designed to make people who are going to fail to finish give up earlier in the ride and therefore be less of a danger to themselves and others by pushing on for longer than they should.
Taking an easy example, Brest is halfway through the ride and therefore a flat average speed would give 45h to get there. If a typical rider just scrapes in to Brest in 44h58m then with just a couple of minutes faffing they'll be in a position where they'll need to completely the second half of the ride faster than they completed the first, which is highly unlikely with an extra 600km in their legs. By forcing them to get there in 43h or so they're guaranteed to have some time in hand on a flat schedule - they could rest for 3 hours and still have more time to finish than they'd taken so far.
Using an argument of extremes, giving people up until 89h to get to Brest would be silly, as that would probably lead to a greater DNF rate than normal as people took too long over the first half. Similarly, only giving people 24h to get to Brest would lead to a huge DNF rate (if the intermediate control closing times were enforced). Therefore the sensible point is somewhere in between these two extremes. I'd guess that ACP believe that it is best set lower than 45h but above 40h and so they've settled on whatever they use now. Setting it at 45h or above does not make sense as the majority of people are not going to be negative splitting PBP.