Proposing a debugging method for this bug
First, thank you for spending time working on the crash at VT switch on
AMD64: fixing this bug is really important for the g-i project.
I've just cloned your original BR as bug 396520  to separate from the
french keyboard issue, retitled it, assigned to the directfb package and
merged with one by davide.
I've cc'ed d-boot this time, but feel free to drop them in future replies.
I've looked at you strace log and i think the best way to proceed is to
check wether the crash can be reproduced or not also on a regular Debian
system, where debugging is easier due to availability of gdb etc.
So, i'm now asking you to install the libdirectfb-0.9-25 package on your
regular debian system: this will allow you to run DFB applications from
it (you'll need to enable framebuffer support at boot time by passing
"video=vesa vga=788" to the kernel).
Now, please compile and install Directfb examples  on your system and
run each df_* application in turn.
While an application is running, please switch to vt1 (ctrl-alt-f1) and
then back to vt7 to see if you can crash your AMD64 system with a pure
DirectFB (no gtk) application.
In the case you should succeed in crashing the application and the
console should become unresonding, you can connect from ssh and issue a
"chvt 7" to get back to X.