Your C++ compiler already implements a solution to that called short string optimization. Strings start out as small byte buffers that can be easily be passed around. When they grow beyond that, the fixed buffer is swapped out for pointer to another allocation on the heap. There's no (immediate) reason that allocation has to come from a direct call to the system allocator though, and it usually doesn't. It can just as easily come from an allocation pool that was initialized at startup.

Even if you needed to hardcode upper size limits, which your compiler already does to some extent (the C/C++ standards anticipate this by setting minimum limits for certain things like string length), you wouldn't actually pay the full price on most systems because of overcommit. There are other downsides to this depending on implementation details like how you reclaim memory and spawn compiler processes, so I'm not suggesting it as a good idea. It's just possible.