Four... times... four... tiles? O.o'
Yeah, I can see how, like, 15360 calls per 640x384 screen can increase render time. O.o'
Also, what is the largest screen size you're going to use? I ask because I wonder whether single BitmapData could cover it all, thus making it possible to draw all these tiley thingies at the very beginning and then only copying the visible fragment onto another bitmap, which would be then displayed during Drawing cycle. Should be this done in the most crude way (as in, the whole screen-sized fragment would be drawn on the "display" bitmap, then that bitmap would be drawn again in Drawing cycle), it should take 2-3 screen drawings per cycle.
Also, I've done a little test, and on my machine drawing 640x360 image 25 times (which is supposed to represent 24-25 screen drawings per cycle) takes about 12-13 ms to render at worst, with both player and browser checked. If you want to check how it runs on your machine, you can get "Efficiency Test" behavior on Forge I just sent. ^^
(and what type of game is that, anyway? There's a huge difference between tile-based Sokoban-like collision and shooter-like pixel/rectangle precise collision ^^')