I'm creating some glTF files, and there aren't many glTF importers yet, so it's possible I may simply be creating my files incorrectly. But I think the Three.js glTF loader isn't handling normalized data for vertex colors correctly.
Here is a file that I'm trying to load in three.js:
http://www.user00.com/threejsnotnormalized.js
In it, I'm creating a GLB with a triangle where the vertex colors at the corners are:
0xFF0000FF, 0xFF0100FF, 0xFF000032
The colors are stored as 8-bit unsigned bytes.
The result I'm seeing is below:

It looks like the 0x01 in the green channel of the 2nd point is being converted to a 1.0 instead of 1.0/255. I see other consistencies when using the vertex colors 0xFF0000FF, 0xFFFF00FF, 0xFF0000FF that are consistent with 0xFF being interpreted as 255 instead of as 1.0.
I think that when I set {"normalized":true} for the accessors, then the data should be normalized, but it seems that three.js isn't doing that.
You can edit for small-test.
http://jsfiddle.net/akmcv7Lh/ (current revision)
http://jsfiddle.net/hw9rcLL8/ (dev)
Haswell integrated graphics
Thanks for reporting.
GLTFLoader doesn't support accessor.normalized yet.
We should let it support.
https://github.com/KhronosGroup/glTF/tree/master/specification/2.0#accessornormalized
Update: I'm working on it...
Thanks @takahirox! Let me know if I should take a look at anything also. :)
Almost done in the local branch.
But I don't have any accessor.normalized file and can't test well enough other than threejsnotnormalized.js.
Any other accessor.normalized files?
Not that I'm aware of. That might be a good request for the https://github.com/bghgary/glTF-Asset-Generator project.
Here is the working repository
Seems it's working for threejsnotnormalized.js but is not tested well enough yet.
I only just saw this but there is an optional normalized argument to THREE.BufferAttribute: https://threejs.org/docs/#api/core/BufferAttribute, which passes into WebGL vertexAttribPointer() call.
Oops, I wasn't aware of that. I'll look into it later...
And also I had missed Related WebGL functions: vertexAttribPointer() normalized parameter in glTF accessor specification...
Updated.
Seems like actually we should use normalized argument.
Nice! That seems more convenient yes 馃槃
That might be a good request for the https://github.com/bghgary/glTF-Asset-Generator project.
Yup. And requesting to https://github.com/KhronosGroup/glTF-Sample-Models seems also nice.
Most helpful comment
Thanks for reporting.
GLTFLoaderdoesn't supportaccessor.normalizedyet.We should let it support.
https://github.com/KhronosGroup/glTF/tree/master/specification/2.0#accessornormalized
Update: I'm working on it...