i apologize. of course blitting cannot beat GPU in pure head to head comparison.
(a 1024 x 768 canvas can only "blit" with 25 Frames/sec on a Motorola Xoom).
my comment just came out of frustration. i did a game similar "bubble breaker" years ago for the sony xperia x1. similar setup ... 800x480 background, 130 unanimated bubbles. all bubbles are drawn on canvas and if i needed to move some of them i just did a redraw of the moving bubbles. of course the framerate was not really good but "ok". even a "non blitting" setup with 1 background bitmap and 130 "normal" flash sprites was ok.
years later we have faster tablets, better GPUs, molehill and GPU based 2D APIs and adobe promising wonders. of course someone would expect to run a simple "bubble-breaker" setup with 60FPS on a tablet using a GPU based framework.
so i did a setup in starling and for comparison in ND2D. the results are pretty much the same. rendering 130 sprites with 1 background sprite performs pretty much equal in both frameworks with 45-47 fps on my XOOM.
of course this is faster than any old-school approach but it is not really as fast as one would expect from all the fuzz adobe is making.
i really _do_ think starling is a step forward for most game-projects.
unfortunately in my 2 cases (my first try to port my boulderdash clone and this simple bubble-breaker "port") there was always something in the way. first it was the missing feature to update a texture efficiently during runtime ... now it is just the number of sprites.
on a sidenote: last friday friend of mine did the same setup in unity with around 100 real "spheres" and a big animated, shaded and rotating skull-mesh (instead of a background) and it did run with 60fps on a ipad2.
as always the grass seems to be greener on the other side 
have fun,
QuadWord
btw: i use BlendMode.NONE and TextureSmoothing.NONE for backgrounds 
