Yo-Yo

Yes, and it was done intentionally. We still have no confirmed data on whether trim schedule block is operative during DIRECT MODE, but we can make an educated guess, that it zeroed, using several facts:
DIRECT is an emergency mode for catastrophic SDU failure while scheduler signal can be false too. Using this signal deadly connected to the system seems to be very optimistic.;
the scheduler can be switched off or zeroed by means, for example, of switching of its supply.
this scheduler is a mean to mimic natural positive velocity-stable trim characteristics when SDU is operative and maintain neutral velocity stability within the whole range of Mach number.
all sources mention "direct link using separate electric supply" but no source mentions this scheduler active.

But if some different and confirmed information is available - things can be changed.

 

Flow separation IS a function of Mach.

It's just because of cruel physics... If you have asymmetric force (thrust) applied it creates a moment. This moment can be counteracted with a pure moment (if you, for example,  start to spool up a giant flywheel at the vertical axis :D) or with a decentered force - a sideforce of the fixed tailwheel you do not have actually.  The braking force of a tire right under the engine could eliminate this moment but the only small detail spoils all - the braking force must be exactly equal to the thrust eliminating it as well. 

The speed depends on radiator scoop position as well, so the max speed depends even at the pre-history of your flight? because the scoop position is not a determoned fuction of coolant temperature. It will stop in different positions as the coolant temperature is in limits. I. e. if your limits are T1 and T2, the scoop can stop finally at any position where heat balance is obtained for the temperature between T1 and T2. The scoop position as well as engine power and compressed air temperature affect charge temperature. But there is exhaust thrust that decreased with rpm decreasing. A lot of factors, indeed. By the way, the blade tip Mach is about 0.92, so the losses starts to rise very fast in this region.
For example, starting at ISA SL at 570 gives you this kind of excessive specific power log.
Three charts are a bit overlapped time history of 3000-2650-3000 profile.
So, reducing rpm you can add few mph, but this overboost condition can be dangerous. And the effect itself lays within 1%, so less than real world measuring ability.