Edited outtake from a now moved post:
Since basic 5-year-old Tesla rear motors seem up for the job of powering this car (to a peak of 10,000 Nm wheel torque using required gear reduction of ~5:1 to achieve 250mph around peak rpm) using three of them combined, unless they want exceptional efficiency and/or long burst performance, batteries will be a tiny factor only towards total car cost. For sure by 2020, 2170's will cost Tesla under $20K for 200 kWh, 10% or less of sale price. As also in fully loaded Model 3's, by the way, $8K in cells towards a $80K car). Let alone Model S/X P100D if that's possible with 2170's (I'm not sure the same power can be drawn for the same amount of time), $10K in cells for a $150K car.
Batteries becoming such a relative afterthought in terms of production cost management, if by 2020 Tesla were able to purchase new batteries of higher gravimetric energy density (say, double), at triple or quatruple the cost, that'd be TOTALLY worth it for he Roadster. It's supposed to be a performance car after all, not an industrial Power Pack stuck to the proverbial Tesla skateboard? Absolutely no need for bottom of the barrel ultra cheap ass 2170's to make up the 200 kWh. For now of course, there are only vague announcements (based on theorerical or lab research) of batteries offering such ballpark densities.
And even if they always intended to equip the $200K Roadster with >10,000 2170 cells, once a true battery upgrade becomes avialable, it will be an easy cell. Who doesn't want to have a sports car that weighs 500-600 kg less than the bulky original? See what people pay extra for a bit of carbon trim that doesn't weigh any less.
I strongly hope the tech will become available, at any price, to see double or triples densities versus today. It will eventually come down inprice, unless the key cost driver is something like for instance silver which is in relatively tight supply (if you need a lot of it).