Automatically extending the charter might lead to a large charter later on. As the "resolution" of the patchwork normative framework grows (i.e. many specific principles), the amount of compute necessary to evaluate the suitability of different actions based on the charter might also grow, potentially limiting capabilities in an economically-undesirable way.
However, assuming the deontic array really is effective in a fundamental way, having a control knob over the intensity of its constraints might be a feature, not a bug. One might imagine a deployment strategy where the amount of compute invested in deontic prudence would be loosely proportional to the model's capabilities and scale at any given moment. Only a model at a certain capability level might be able to game a given charter, so it seems sensible to invest the associated compute then, perhaps somewhat conservatively early.
In a sense, the alignment tax would be framed as a percentage commission of the whole project, growing over time in an absolute sense but remaining relatively tiny.