I made some tests using the tilemap object and discovered that (for me) most seams are caused by tile textures wrapping around themselves.
Here you can download the capx:
https://dl.dropboxusercontent.com/u/7871870/construct/tilemap%20seams/bug-tilemap-seams-02.capx
On layout 1 I used the image below as a tilemap, then placed a single "red-slope" tile in the layout and scrolled to a non integer position.
(checked parts being transparent)
Below is the result, here enhanced to be easier to see:
Clearly no other color is bleeding to this tile, only the red from the same tile is wrapping around and causing the seams:
So this indicated that the problem must reside in texture wrapping, and not tile bleeding.
On layout 2 I tested how a repeating tile behaves when placed as a tilemap, a pre-rendered sprite, or a tiled background, when both the scroll position and layout scale have non integer values. This is the result:
Surprisingly only the sprite is not wrapping and not causing seams.
I know there's a wrap parameter used to specify how the texture is sampled when the uv coordinates wrap in WebGL.
If set to "repeat" it wraps the texture around, and if set to "clamp_to_edge" it repeats the edges.
I'm not sure of C2 internals, but what I believe is happening is that C2 uses "repeat" to help generate the tiling textures, like for repeated tilemap regions and tiled backgrounds, and uses "clamp_to_edge" only when drawing single textures like sprites.
So I speculate that when C2 draws a repeated tilemap region, it draws directly to the scaled screen size using "repeat", which in non integer sizes generates wrapping seams due to floating coordinates:
If that's the case, then maybe it's possible to fix the problem by first drawing the tiling to a buffer texture using "repeat", in a 1:1 scale (or the next fraction of 2 proportional to the screen coordinates, ex 1/2, 1/4, 1/8...) so no wrapping seams are generated, and only then draw the buffer to the screen size using "clamp_to_edge", like this:
If that happens to work then it may be possible to keep the tilemap rendering optimizations at least for WebGL, don't know about Canvas though.
And if there's still seams caused by gaps between tiles, maybe it's possible to add an option to stretch the buffer texture to extend one (or half?) pixel to each side in screen coordinates when drawing, so there's a small overlap between each tiled region. For high-resolution tiles and displays the stretch should be almost unnoticeable and might solve the seams problem.