Ya, the value of collision meshes is so strong that I am surprised that this was not already done.
This is almost certainly because the core of the engine appears to scare everybody.
There has been very little work done in the guts of engine recently - this isn't scientific, but the svn history of modelcollide.cpp has only 5 commits since 16th March 2008.
- The whole of the model folder has 64 commits, mostly animation, table and SEXP related.
Obviously, none of this gives any indication of how significant or difficult any of these changes was, but it is indicative of how little work has been done there recently.
(There aren't any valid metrics for importance or difficulty. A one-character bugfix could be amazingly hard and important, fifty lines of code could be trivial)
As to why - this is probably documentation.
A tiny change in the core of the engine is likely to affect almost everything - this makes it extremely dangerous to go in without knowing exactly what you're doing.