>
> it: I'd be much more likely to run an experimental gateway in another
> process on a node than experimental p2p code inside my production
> bitcoinds themselves.
>

Yes, it's certainly better to do that during the development phase. However
if it does turn out to be good and valuable then it'd eventually need to be
integrated or rewritten into Core anyway, lest we accidentally increase the
setup cost of running a node and end up with a two-tier network. And if the
code will eventually want to be merged into Core anyway, it might as well
be implemented into it directly, perhaps behind a switch that can disable
those codepaths if something goes wrong.

So I think the tradeoffs here are rather complicated and subtle.