One problem I could see is that even if you could pull the lexical values, you might have serialization problems. For instance, if two closures shared a ref and upon re-animating, getting them to both point back to one value instead of two independant copies. I'm sure you could do it but I would think it might be tricky in subtle ways.