Quick research into it tells me that OpenGL ES, although written against the OpenGL 1.5, 2.0, and 3.0 standards, are not full implementations of them, especially the lack of glBegin() and glEnd() (and I'm pretty sure that FSO uses a large portion of glBegin() and glEnd() 's as its fallback graphic engine). Ironically enough, the "legacy cruft" that Nuke mentioned would seem to be exactly what an embedded graphics platform would need.
I also noticed that OpenGL ES provides a duplicate set of functions that take fixed-point numbers instead of float32's, as well as come across documentation that suggests that some OS's convert the fixed-point numbers into float32's so it can be passed to the embedded GPU.
Now, unless the fixed-point GPU's are specifically fabricated against the GLfixed type, I would have to say that there is absolutely no gain to be had from this method, especially when they're going to be converted to and from float32's. The end result is that you may end up with an app that gets half to a quarter of the frame rate of that of an app written against the full OpenGL standard.
I see that you've managed to run an instance of FSO, and that tells me that you're probably running it on a machine that supports float32's, and if not, it's already doing the fixed->float->fixed madness.
P.S. I think it would also help if we knew exactly what kind of fixed point representation it's messing with...
ShivanSpS, which tablet did you run that on?