You have to keep in mind that with these examples the runtimes of both approaches are basically already really fast. Probably a good portion of the time is returning the result set from the database server to the application server and not DB processing anyway.
This actually brings up a good point. These are basically a single query operation which is never really a good use case for code pushdown. Its not just about pushing application logic down into the database, but also data flow. If you have results of a few queries that then is needed in later logic and you can do the complete block in AMDP; you avoid transferring the intermediate data to the application server. This is where you start seeing some gains. The larger the block of code (and subsequently the less data round trips to/from the DB) the better the improvement in the runtime.
I've also seen the case where people write lots of small AMDP methods - basically one method per SQLScript operation. This is a bad pattern as well. You aren't allowing HANA to parallelize the code nor are you eliminating the data transfer to the application server per operation. So there are certainly anti-patterns of usage of AMDP that can result in lower performance.