10 million records in 3 hours = 925 records per second.
To achieve some sort of parallel processing you will need to split the SQL message in the pipeline, process in multiple orchestrations and append the output records into a file. The receive port and send port will be single threaded, but the orchestrations will multi-thread.
If you require the order of records to be preserved, then you will need to single thread the entire process.
Personally I would look at using SSIS, Biztalk introduces a lot of overhead that you do not need for this process.