It would help if he defined what "subpixel accurate" means. I can't figure out why you'd care about subpixels if you're producing a bitmap image? Or is he doing something like cleartype but not antialiased?
With subpixel precision (e.g. you do all the math at a higher precision than the framebuffer resolution, and preserve the fractional part of pixel coordinates in all computations until you actually write to the framebuffer), the outlines remain stable even for small movements. IIRC Quake (with the software renderer) was one of the first games which paid proper attention to subpixel accuracy, and they showed this off with a very slight camera movement at the result screen after a multiplayer match. Triangle edges were properly 'crawling' instead of being all jittery.
For example, the standard algorithm would always generate lines that are symmetrical:
####
####
But the modified version lets you position the endpoints different, so that you'd get: ###
#####
or ##
######
etc 0 #
#######
##
######
###
#####
3 ####
####
#####
###
######
##
6 #######
#