rotate_render.py

import bpy
import bmesh
import time
import math

path = "/home/zeffii/Desktop/tmpo5/"
context = bpy.context

for area in context.screen.areas:
    if area.type == "VIEW_3D":
        ctx = { "window": context.window, "area": area, "region":None}
        break

US = lambda: bpy.ops.wm.redraw_timer(type='DRAW_WIN_SWAP', iterations=1)

obj = bpy.data.objects['Suzanne']
bpy.context.scene.objects.active = obj  # not only active but also selects.
bpy.ops.object.mode_set(mode='OBJECT')

for i in range(9):
    obj.rotation_euler = ((0.0, 0.0, math.pi*i/4.0))

    US()

    for k, f in enumerate(obj.data.polygons): 
        f.material_index = 1 if k%(i+2) else 0

    time.sleep(2.0)
    file = "ao2_" + str(1000+i) + ".png"
    bpy.ops.screen.screenshot(ctx, filepath=path+file, full=False)

there's no need to really use bmesh just to modify mesh data and face materials, it's worth seeing if by not using bmesh or switching object/edit modes -- perhaps something is different.

• If not, then at least you know that's not the issue, but you can focus on less code.
• if it does make a difference, it narrows down the area to look at.

The other difference with doing screenshots manually is that blender does a bunch more updates when you move your mouse around the UI, i would consider sticking a few more calls to US() .. maybe even directly before taking the screenshot

 US()
 bpy.ops.screen.screenshot(ctx, filepath=path+file, full=False)