I first present these recurrences and how they map to pentagons:
In practice I do not just subdivide; I also expand by φ = (1 + √5) / 2 each time and rotate around the point which the first triangle and the subtriangle created at its base have in common by 2π/5 radians. This guarantees that each subdivision merely adds to, without changing, the previously generated area. The result is a Penrose tiling:
Penrose tilings are peculiar in that they exhibit no translational symmetry (that is, they never repeat) but they might have 5-fold rotational symmetry. If I create ten triangles point-to-point as my seed (or, equivalently, ten gnomons) then I will create two different rotationally symmetric tilings, one with a star in the center and one with a pentagon in the center, depending on the number of expansion steps I use. It also looks as though the tile above has rotational symmetry, as is made more obvious if I superimpose the darts (i.e., the gnomons) on top of the tiles
Much else could be said; the the coloring makes things which look like, but are not, Ammann bars appear where a single straight line appears to interect only one color of pentagon. However, if looked at closely, there are some few pentagons of the other color on each stripe (if you look far enough; I have not proven this, but empirically any stripe I find if I expand the view I see broken). There are also lines that never pass through stars and other odd features.
Note that in any non-repeating tile no finite set of observations is enough to locate oneself. In general, given any k-radius sub-image of the tile, an identical subimage can be found within a distance of roughly 2k, though there are also other constraints; for example, two stars of the same orientation can never appear closer than do the five surroundign the central star of the images on this page.
Finally, I offer a cool picture (but big; nearly 1Mb) of an infinite Penrose-style pentagonal tiling taken with a spherical lense (implemented as an OpenGL vertex shader).