The backend made some improvements to the performance of time-to-first-byte for query results. I'm working on implementing these changes here: googleapis/python-bigquery#362

pandas-gbq will not be able to take advantage of many of these performance improvements unless it uses the return value from QueryJob.result() or even QueryJob.to_dataframe() directly.

• Use QueryJob.to_dataframe() (remove any manual list_rows() on destination table calls if still present).
• Use Client.load_table_from_dataframe() (Done in feat: to_gbq uses Parquet by default, use api_method="load_csv" for old behavior #413)

Related: #149