It is indispensible, and I think that speeding it up will be not just within feasibility, but have the appealing effect of speeding up the general facilities of the language!
But the next great challenge I'd like to be able to prototype is to be able to run on top of an abstracted pipe vs. marching along directly in memory. If we could do that, we could run on network data...or be processing a continuously feeding pipe from a CALL session that is being scripted and reacted to.
It might be another one of those "you get what you pay for" situations with the combinators. If the combinator only accepts ANY-SERIES! input and not PORT!, then it might be more limited. But we'd want the stock combinators to be able to run on either.
Point being: still more prototyping work to be done before the optimizing!