A bad gridd fitting happens if two stems have a common boundary, and that boundary has a lower alignment strenth, and each stem induces a different coordinate. Happens with KozukaB-ILEmbed.pdf when rendering wih patch7 of the bug 687654.
Created attachment 2030 [details] KozukaB-ILEmbed-1.pdf A simplified test case. Horizontal stems in 5 and 7 get different width due to an alignment conflict in 5.
Created attachment 2031 [details] 5.bmp.zip A visual trace diagram which demonstrates the effect. The upper horisontal arm is controlled by hint[2] and hint[3]. The lower boundary gets a conflict, and the upper coordinate is being choosen randomly. Must choose those, which corresponds to the longer stem.
Created attachment 2032 [details] patch.txt A suggested patch is being tested.
Patch to HEAD : http://ghostscript.com/pipermail/gs-cvs/2006-February/006347.html This patch also fixes the bug 687654.