I am running a very simple SQL script within a notebook, using an X-Small SQL Serverless warehouse (that is already running). The execution time is different depending on how it's run:

• 4s if run interactively (and through SQL editor)

• 26s if run within a workflow as a notebook task

If you look at the query history screenshot below you see the individual queries are taking the same amount of time to execute, but for some reason there's a few seconds delay between end of one query and start of the next query (line) when running through workflow vs interactively (both using SQL serverless warehouse). I tried both current and preview channel and they both behave the same way

I even tried using an all purpose compute cluster (Single node: Standard_DS3_v2 · DBR: 15.3) and the times were consistent when running in a workflow vs interactively.

Script

declare or replace start_time timestamp = current_timestamp();
declare or replace end_time timestamp = current_timestamp();
declare or replace v STRUCT<SOURCE_TBL_MAX_TIMESTAMP STRING, TARGET_TBL_MAX_TIMESTAMP STRING, SOURCE_MAX_TIMESTAMP TIMESTAMP, TARGET_MAX_TIMESTAMP TIMESTAMP, MIN_TIME TIMESTAMP, MAX_TIME TIMESTAMP, MAX_INTERVAL_TIME TIMESTAMP, TABLE_SOURCE STRING, TABLE_TARGET STRING, MERGE_STATEMENT STRING, INTERVAL_COLUMN_NAME STRING, STAGING_TABLE_NAME STRING, STAGING_TABLE_CREATE_STATEMENT STRING, FULL_STAGING_TABLE_NAME STRING>;
select current_timestamp();
select v;
select current_timestamp();
select timestampdiff(MILLISECOND  , start_time, current_timestamp()) as total_time_elapsed_ms;

Screenshots 

Query History

Further tests

Using a SQL file workflow task, running on X-Small SQL serverless, the same script executes as expected (faster than notebook task):

Not sure why the type of workflow task would impact the execution of the SQL script on a serverless warehouse.