Of course, in practice, implementing an algorithm that performs well in experiments will only be slightly polished before putting it into production. Even in spite of the relative low cost of EC2 hosting, developers trying to maximize their ROI might find that the equation balances in their favor by maximizing efficiency on each core and decreasing the amount of memory that can be used. I admittedly don't know a lot about scaling experimental code into enterprise-level code, but I do know that at some point, it must reduce to a cost-benefit equation, where the costs of EC2 hosting is compared against the costs of development/debugging time in languages like C++. This, of course, reduces to the specifics of your business needs.