Does "unused" metadata have impactful performance implications?

We are exploring breaking out our existing application and using multiple pods in our environment to help improve performance and run multiple models at the same time.

Each pod only needs a subset of the members in a couple of dimensions (e.g Parent A but not parent B, C and D). All of our existing allocations have already been set to only use these relevant parents to help improve performance.

However if we hypothetically had an application where all allocations only reference say ~20% of the members in large dimensions (1000 - 100k members depending on the dimension), will the run time of those allocations be signifcantly faster if those dimensions did not have those extra members?