Performance is important for many users of PyPDF2. For this reason we should introduce some kind of performance testing.

I would approach it like this:

1. Identify typical workloads where performance might be / become an issue.
2. Create a benchmark. This includes scenarios (e.g. doing something with one big file vs doing something often)
3. Define a baseline scenario for which we measure the time. All other times are just relative to the baseline (e.g. 1.78x the baseline). The baseline should be something that is unlikely to change a lot. By this relative approach we hopefully get rid of machine differences