> Is it worth implementing a cache layer in a system?

This isn't a "Tech Debt" thing. This is a feature/scoping thing. Does my system need to handle 1M concurrent users? If so, support that. If not, don't support it.

But we don't need the absurd concept of "Tech Debt" for effective analysis. We can be specific. We can be clear. Does my system need to support low latency in the face of 1M concurrent users? Yes or no. Now proceed.

"Tech Debt" is such a ridiculous dodge to otherwise avoid real analysis, real work, real clarity.