When I wrote about \mathsurround I had also in mind another limitation, but this one seems much more dificult to change. It is the fact that we may only want the mathsurround glue to be inserted if the math formula would otherwise still be adjacent to a glue or a kern, not counting intermediate whatsits or penalties; or better, to have a \shortmathsurround for the other cases. It is even delicate to formaly define the rules we want to follow. For the moment I think it is enough if \mathsurround is changed to be a skip. However I will not forget about this idea. If muthsurround exists it should be possible to have it behave the way we would like (as authors). When I have looked at the tex sources and learned more about the internals I will try to look about it, but it is likely that this year I will have very few time.