The unicode standard dicusses a syntax for identifiers here [1]. Note  
that determining if two identifiers are equal needs to invoke the  
unicode normalization machinery, it is not just a matter of knowing  
what is a letter and a symbol. That being said, I really don't see  
unicode identifiers as a must have feature, au contraire.

Regarding the use of math symbols, I think Alain is right, it should  
be addressed as a presentation issue. Actually that's the way Fortress  
[3] handles it, as a rendering issue.

Speaking of using UTF-8 in source files, it was recently pointed out  
to me that this is not a problem as long as you use only ASCII  
identifiers. The advantage is that you can have UTF-8 string literals  
in your source code.

Best,

Daniel

[1] http://www.unicode.org/reports/tr31/
[2] http://research.sun.com/projects/plrg/Publications/fortress.1.0.pdf