One thing I'd like to see lwjgl do is hide the fact that these nio buffers are used and implement the calls to OpenGL as the OpenGL documentation suggests i.e. by passing in Float rather than pointers to a FloaftBuffer to the various calls. This would certainly make porting existing OpenGL applications written in other languages a whole lot easier.
While I agree with you there are 2 reasons why I don't think lwjgl should go in this direction:
1) passing in floats mean taking stuff out of the Java stack and moving it into native memory used by the OpenGL driver. Pain and suffering performance wise - especially on architectures where java byte ordering is different than system byte ordering.
2) It would be a maintenance nightmare to have to wrap every individual function call and implement all of the various possible data types that could be passed into these functions. To be honest there is very little gained and for people who've written applications where they were managing their own memory in C/C++, its actually more natural to do it this way than to wrap the function calls.
Its not the fist time its come up, but the performance and maintenance implications far outweight the marginal increase in ease of use. If it were possible to have both, I would do it myself
Now if only I could get Apple to release JDK1.4.1 for OSX... I simply refuse to port any more until I can tell whether the bugs are in Java, LWJGL, or my own code!