griatch: please could you attach a small minimal "bad" .ora file and a .jpg of what it's supposed to look like saved from Gimp? I'll need the sample file it to fix this one. Thanks!

My gut feeling from looking at it is that MyPaint might be applying the file gamma incorrectly during processing (oops).

maxy: from your description, the file had no ICC profile embedded to say what its R,G,B[,A] samples mean, nor a chunk to indicate that they're sRGB values with standard gamma compression. Thus we fall back on treating it as a "generic" additive RGB colour space where the R, G and B primary colours the samples express degrees of are (maybe) defined in absolute XY[1] terms along with the white point[2], and (separately maybe) there's a gamma correction curve. lcms2 can handle all 3 cases, but probably only if we feed it the right data...

[1] http://en.wikipedia.org/wiki/CIE_XYZ
[2] http://search.cpan.org/search?query=Graphics%3A%3AColorObject&mode=module is a really good start point for how this sort of colour munging works.

Saving an image in GIMP 2.6.11 using the latest ora-plugin works just fine. You can also re-load the file in GIMP without any problems. But when loading that .ora file into MyPaint-GIT, it gets a "white shade" applied to it that essentially destroys the image. Saving the file in MyPaint and importing it back into GIMP retains this white shade effect.

Below is a demonstration of what happens. To the left is the original .ora image, saved in GIMP. To the right is what comes back after having loaded and saved that image in MyPaint.

The image is a two-layer image, not different from any other I've used many times previously. I have tried this with single and multi-layer images too, getting the same result.
.
Griatch