Looks quite likely like a driver bug. You could try to restart kwin from a command line after that crash occurred:
kwin --replace &

Anyway, could you please upload the output of:
qdbus org.kde.kwin /KWin supportInformation

and
glxinfo -l

*** Bug 308398 has been marked as a duplicate of this bug. ***

Created attachment 74547 [details]
glxinfo

Created attachment 74548 [details]
qdbus org.kde.kwin /KWin supportInformation

Attached my information. KWin smoothly restarts when it crashes and (in my case) it only crashes when enabling compositing.

What's your (combined?) screen resolution?

Laptop Display: 1920x1080, LCD connected via VGA: 1920x1200

Do you get the same crash with the internal display only?

If only one of them is active, Desktop Effects can be enabled without any problem. Also if the second display is activated afterwards, the Desktop Effects still work. If they are deactivated, they can't be activated again without crashing KWin.

Do you have a smaller external screen so that if you arrange the displays vertically you stay below 2048px in either direction (ie. the external height should be < 969px) or can you try to arrange the displays overlapping to stay in that bounds?

Created attachment 74566 [details]
glxinfo from my lenovo X230

Created attachment 74567 [details]
kwin supportinfo from my leovo X230

(In reply to comment #10)
> Do you have a smaller external screen so that if you arrange the displays
> vertically you stay below 2048px in either direction (ie. the external
> height should be < 969px) or can you  oo arrange the displays overlapping
> to stay in that bounds?

It turns out if I first turn off Laptopo screen and then turn on extmonitor in lower resolution(1920x1440) kwin does not crash then I switch to highest res (2560x1440) afterwards.

like this litle script:

if [ "$1" == "--lpoff" ]
  then
    xrandr --output LVDS1 --off

    xrandr --output DP1 --mode 1920x1440
    xrandr --output DP1 --mode 2560x1440
    exit 0
fi

(In reply to comment #10)
> Do you have a smaller external screen so that if you arrange the displays
> vertically you stay below 2048px in either direction (ie. the external
> height should be < 969px) or can you  oo arrange the displays overlapping
> to stay in that bounds?

It turns out if I first turn off Laptopo screen and then turn on extmonitor in lower resolution(1920x1440) kwin does not crash then I switch to highest res (2560x1440) afterwards and it works.

like this little script:

if [ "$1" == "--lpoff" ]
  then
    xrandr --output LVDS1 --off
    #if I remove this line kwin crash when call this script with --lpoff
    xrandr --output DP1 --mode 1920x1440
    # the res I want
    xrandr --output DP1 --mode 2560x1440
    exit 0
fi

#However turning on both screen does not work with Compositor it seems...
if [ "$1" == "--both" ]
then
  xrandr --output LVDS1 --mode 1366x768
  xrandr --output DP1 --mode 1920x1440
  xrandr --output DP1 --mode 2560x1440 --left-of LVDS1 --primary
  exit 0
fi

> #However turning on both screen does not work with Compositor it seems...
Does it work with 1920x1440 on the ext screen or does this

xrandr --output DP1 --mode 2560x1440 --pos 0x0 --primary
xrandr --output LVDS1 --mode 1366x768 --pos 683x0

work? (makes LVDS1 extend DP1 by 1px to the right)

It will likely be size driven (and in that case also a driver bug, since the GPU or at least MESA propagates:

GL_MAX_TEXTURE_SIZE = 8192
GL_MAX_VIEWPORT_DIMS = 16384, 16384
GL_MAX_RECTANGLE_TEXTURE_SIZE_NV = 4096

and you're not near any of those limitations.)

random maybe unrelated note: we currently have quite a lot of various reported issues around the combination of Kubuntu 12.10, Mesa 9 and Intel. It seems to be all Ivybridge. Somehow I'd say they are all related and I'd love to figure that one out before the Kubuntu release

As a reference for example Bug #308245

(In reply to comment #15)
> > #However turning on both screen does not work with Compositor it seems...
> Does it work with 1920x1440 on the ext screen or does this
> 
> xrandr --output DP1 --mode 2560x1440 --pos 0x0 --primary
> xrandr --output LVDS1 --mode 1366x768 --pos 683x0
> 
> work? (makes LVDS1 extend DP1 by 1px to the right)
> 
> It will likely be size driven (and in that case also a driver bug, since the
> GPU or at least MESA propagates:
> 
> GL_MAX_TEXTURE_SIZE = 8192
> GL_MAX_VIEWPORT_DIMS = 16384, 16384
> GL_MAX_RECTANGLE_TEXTURE_SIZE_NV = 4096
> 
> and you're not near any of those limitations.)

(In reply to comment #15)
> > #However turning on both screen does not work with Compositor it seems...
> Does it work with 1920x1440 on the ext screen or does this
> 
> xrandr --output DP1 --mode 2560x1440 --pos 0x0 --primary
> xrandr --output LVDS1 --mode 1366x768 --pos 683x0
> 
> work? (makes LVDS1 extend DP1 by 1px to the right)
> 
> It will likely be size driven (and in that case also a driver bug, since the
> GPU or at least MESA propagates:
> 
> GL_MAX_TEXTURE_SIZE = 8192
> GL_MAX_VIEWPORT_DIMS = 16384, 16384
> GL_MAX_RECTANGLE_TEXTURE_SIZE_NV = 4096
> 
> and you're not near any of those limitations.)

Buserror:

DP1 1920x1440
LVDS1 1380x768 

Below works (except the position of LVDS is confusing...) 

xrandr --output DP1 --mode 2560x1440 --pos 0x0 --primary
xrandr --output LVDS1 --mode 1366x768 --pos 683x0

What about this:
xrandr --output DP1 --mode 2560x1440 --pos 0x0 --primary
xrandr --output LVDS1 --mode 1366x768 --pos 1706x0

and in copmarism this:
xrandr --output DP1 --mode 2560x1440 --pos 0x0 --primary
xrandr --output LVDS1 --mode 1366x768 --pos 1707x0

About Commen 18:

The first crashed kwin BUT this time effect was still active after restart of kwin.... (No need to go to the effects setting to enable composite)

Second worked also and didn't crash kwin.

Also tried some more tweeking by start around --pos 2000x0 increment pos with step around 100 but when reach around >2400 I step increment only by 10 pix until I reached 2550 when a then changed to 2560 it crashed....

then I changed res back to 2550 and tried to enable desktop effects and do kwin --replace but then kwin crashed hmm...

Means it was likely some different, unrelated patch (Ubuntu had a bad libxrandr version)

However, there seems no "magic hidden border" like 2048+1024 which causes this. There's just some memory corruption error on the higher areas.

If you move the screens on top of each other, can you offset the y of the smaller one to 3328 (so you'll get 3328+768=4096)?

Created attachment 74628 [details]
New crash information added by DrKonqi

kwin (4.9.2) on KDE Platform 4.9.2 using Qt 4.8.3

- What I was doing when the application crashed:
I was connecting an external 27" monitor to my 14" laptop via VGA.  
- Unusual behavior I noticed:
The first time I connected the monitor, kwin didn't crash, the second time did. 
Both times the computer became slow. When disconnecting the external monitor the laptop was thinking to be larger than it actualy is, so I have a good part of my application windows outside the visible area (I've taken a screenshot if needed)
As already stated logging out and in again reenable effects and the normal size of the screen.
I'm also on Intel Ivy Bridge with integrated graphics.

-- Backtrace (Reduced):
#6  __memset_sse2 () at ../sysdeps/x86_64/multiarch/../memset.S:908
#7  0x00007f51ea70cde6 in memset (__len=<optimized out>, __ch=255, __dest=<optimized out>) at /usr/include/x86_64-linux-gnu/bits/string3.h:85
#8  intel_miptree_alloc_mcs (intel=intel@entry=0x20663e0, mt=0x22648f0, num_samples=num_samples@entry=8) at intel_mipmap_tree.c:823
#9  0x00007f51ea70d0cd in intel_miptree_create_for_renderbuffer (intel=intel@entry=0x20663e0, format=format@entry=MESA_FORMAT_ARGB8888, width=<optimized out>, height=<optimized out>, num_samples=num_samples@entry=8) at intel_mipmap_tree.c:480
#10 0x00007f51ea70ddf7 in intel_miptree_create_for_dri2_buffer (intel=intel@entry=0x20663e0, dri_attachment=1, format=MESA_FORMAT_ARGB8888, num_samples=num_samples@entry=8, region=0x20f0f70) at intel_mipmap_tree.c:372

correlates with apparent memory curruption of bug #308245
same system, same OS, same MESA 9.0

can anyone reproduce this bug on (downgraded) MESA 8.x?

Finally got some time over to downgrade libmesa to 8.0.4 and the problem with crashing kwin is gone!

also  bug #308245 so yes those bugs in combination of mesa 9 seems related atleast onb GMA4000.

There's somewhere a memory corruption and given bug #308245 and other bugs we get around mesa 9.0, it's there - or at least in the 64bit version shipped by ubuntu.

THIS is likely the multisample issue, see bug #308385

*** This bug has been marked as a duplicate of bug 308385 ***

*** Bug 316702 has been marked as a duplicate of this bug. ***

*** Bug 318010 has been marked as a duplicate of this bug. ***