I guess it could be tricky to implement the same semantics as the current pike runtime (refcounting, gc characteristics, the runtime vs compile time type behavior, threading, codecs, etc).
As for the stability argument, I don't consider the current pike runtime unstable at all. Do you? It's old code, tried and true by now.
I think it's more interesting to interface with the .net library, which is extensive and has a well thought through overall design.