We definitely would like to see CompilationTarget -> "C" supported in the cloud down the road. The limiting factor there is that we currently block both child processes (for invoking the C compiler) and LibraryLink (for arbitrary user-generated LibraryFunctions). You can use Parallelization->True now to speed up certain types of Compile expressions.
To your broader question, note that you can develop your code and measure its performance in a notebook prior to deploying it. Also, now that we have a product focused on programming that incentivizes efficient code, we'll have both better avenues for disseminating best practices and an interested audience for this topic.
