(In reply to comment #5)
> Weird. I hope Zack can track this down ... gDumpPaintList=1 would be a good
> start

The display lists don't seem to provide any clues. When the phantom scrollbar is drawn there are a few extra entries for the scrollbar and scrollbarbutton etc, but that is the only difference.

Here is the relevant display list output from a plain trunk build (where the bug occurs).

Painting --- before optimization (dirty 0,0,61440,34800):
SolidColor 0xb0e925b0(Viewport(-1)) (0,0,61440,34800)
Clip (nil)() (0,0,61440,34800)
    CanvasBackground 0xb0e92690(Canvas(html)(-1)) (0,0,61440,34800) uniform
Clip (nil)() (0,0,61440,34800)
    Background 0xb0eb7d10(HTMLScroll(div)(1)) (480,480,6000,6000) opaque uniform
    Background 0xb0eb7dfc(ScrollbarFrame(scrollbar)(-1)) (480,6480,6000,120)
    Border 0xb0eb7dfc(ScrollbarFrame(scrollbar)(-1)) (480,6480,6000,120)
    Background 0xb0ebb040(ButtonBoxFrame(scrollbarbutton)(-1)) (540,6540,840,840)
    Background 0xb0ebb1fc(ButtonBoxFrame(scrollbarbutton)(-1)) (5580,6540,840,840)
    Clip (nil)() (480,480,6000,6000)
        Text 0xb0ebba0c(Text(0)) (480,480,540,1140)
        nsDisplayReflowCount 0xb0ebba0c(Text(0)) (480,480,540,1140)
Painting --- after optimization:
SolidColor 0xb0e925b0(Viewport(-1)) (0,0,61440,34800)
Clip (nil)() (0,0,61440,34800)
    CanvasBackground 0xb0e92690(Canvas(html)(-1)) (0,0,61440,34800) uniform
    Background 0xb0eb7dfc(ScrollbarFrame(scrollbar)(-1)) (480,6480,6000,120)
    Border 0xb0eb7dfc(ScrollbarFrame(scrollbar)(-1)) (480,6480,6000,120)
    Background 0xb0ebb040(ButtonBoxFrame(scrollbarbutton)(-1)) (540,6540,840,840)
    Background 0xb0ebb1fc(ButtonBoxFrame(scrollbarbutton)(-1)) (5580,6540,840,840)
Painting --- before optimization (dirty 480,480,6000,6000):
SolidColor 0xb0e925b0(Viewport(-1)) (0,0,61440,34800)
Clip (nil)() (0,0,61440,34800)
    CanvasBackground 0xb0e92690(Canvas(html)(-1)) (0,0,61440,34800) uniform
Clip (nil)() (0,0,61440,34800)
    Background 0xb0eb7d10(HTMLScroll(div)(1)) (480,480,6000,6000) opaque uniform
    Clip (nil)() (480,480,6000,6000)
        Text 0xb0ebba0c(Text(0)) (480,480,540,1140)
        nsDisplayReflowCount 0xb0ebba0c(Text(0)) (480,480,540,1140)
Painting --- after optimization:
Clip (nil)() (0,0,61440,34800)
    Background 0xb0eb7d10(HTMLScroll(div)(1)) (480,480,6000,6000) opaque uniform
    Clip (nil)() (480,480,6000,6000)
        Text 0xb0ebba0c(Text(0)) (480,480,540,1140)
        nsDisplayReflowCount 0xb0ebba0c(Text(0)) (480,480,540,1140)

And here is the display list from a build with the change in bug 485030 reverted (the bug does not occur).

Painting --- before optimization (dirty 0,0,61440,34800):
SolidColor 0xaff995b0(Viewport(-1)) (0,0,61440,34800)
Clip (nil)() (0,0,61440,34800)
    CanvasBackground 0xaff99690(Canvas(html)(-1)) (0,0,61440,34800) uniform
Clip (nil)() (0,0,61440,34800)
    Background 0xaffcbd10(HTMLScroll(div)(1)) (480,480,6000,6000) opaque uniform
    Clip (nil)() (480,480,6000,6000)
        Text 0xaffd3a0c(Text(0)) (480,480,540,1140)
        nsDisplayReflowCount 0xaffd3a0c(Text(0)) (480,480,540,1140)
Painting --- after optimization:
SolidColor 0xaff995b0(Viewport(-1)) (0,0,61440,34800)
Clip (nil)() (0,0,61440,34800)
    CanvasBackground 0xaff99690(Canvas(html)(-1)) (0,0,61440,34800) uniform
Painting --- before optimization (dirty 480,480,6000,6000):
SolidColor 0xaff995b0(Viewport(-1)) (0,0,61440,34800)
Clip (nil)() (0,0,61440,34800)
    CanvasBackground 0xaff99690(Canvas(html)(-1)) (0,0,61440,34800) uniform
Clip (nil)() (0,0,61440,34800)
    Background 0xaffcbd10(HTMLScroll(div)(1)) (480,480,6000,6000) opaque uniform
    Clip (nil)() (480,480,6000,6000)
        Text 0xaffd3a0c(Text(0)) (480,480,540,1140)
        nsDisplayReflowCount 0xaffd3a0c(Text(0)) (480,480,540,1140)
Painting --- after optimization:
Clip (nil)() (0,0,61440,34800)
    Background 0xaffcbd10(HTMLScroll(div)(1)) (480,480,6000,6000) opaque uniform
    Clip (nil)() (480,480,6000,6000)
        Text 0xaffd3a0c(Text(0)) (480,480,540,1140)
        nsDisplayReflowCount 0xaffd3a0c(Text(0)) (480,480,540,1140)

I'm continuing to look into this...