BSOD 0x1A MEMORY_MANAGEMENT - possibly a regression (I can make it happen only with CarrierReleaser Builds)

0.4.8-RC-25-g11bf8fa gcc dbg

I experienced sporadic BSOD 0x1A MEMORY_MANAGEMENT 3 times with 0.4.8-RC-25-g11bf8fa

I have no clear reproduction steps, but there is a bit of a pattern:
In all 3 cases when it happened, I installed AC97 drivers, then installed another program (e.g. RevoUninstaller or Python 3.4.3) and then rebooted.

In all 3 cases it was when it was almost showing the shell bootup.png (and doing AC97 stuff) when it BSODed:

MmInsertRmap tries to add a second rmap entry for address ...

see:
BSOD0x1A_after_revoUninstall.log
python_ac97_mm_bsod.log

I am pretty sure that I introduced this myself when I partially reverted smileys sxs work in 0.4.8-RC-25-g11bf8fa - < this was a wrong assumption

I can make it happen with carrier releaser builds only
0.4.8RC24
0.4.8RC25
0.4.8RC26

My local Windows builds are not affected.

It seems to be triggered by sxs related code eventually? But it was not introduced by my sxs revertings.