I lost trust of DuckDB's test after https://duckdb.org/2024/12/06/duckdb-tpch-sf100-on-mobile#a-..., they were comparing apples with bananas, well actually Swift app on iPhone vs CLI app in a terminal.
The terminal and CLI app within ran locally on a smartphone, which was the premise of the experiments within the linked post.
They also weren't comparing a Swift app on an iPhone with their Android run, they were comparing both against "... the system in the research paper that originally introduced vectorized query processing[.]"