Answer

• How many connections you have in your workbook
• How the data in each source is related
• The number of records you have in the data
• The types of views you will need to create

• Tables from different data sources can usually be joined using a cross-database join. See Combine tables from different databases.
• Tableau Desktop cannot join published data sources, most extract-only data sources, or cube data sources. You may consider moving the data to another data source, or creating a local copy of a published data source, in order to use a cross-database join, otherwise the data must be blended.
• If the secondary table has a large amount of data then data blending may be faster, because data blending will aggregate the data first.
• Table joins are better when tables have a 1:1 relationship (i.e. there is only one record for each value in the linking fields in each table). If the tables have a 1:many or many:many relationship this creates duplicated data after the join. See Over-counting values or duplicated data
  Note: Duplicated data can be resolved after the table join is created, but the increased amount of data may cause performance issues.
• Data blending has several limitations due to design constraints. For more information about common issues with data blending, see Troubleshoot Data Blending as well as Related Links in the right-hand side of this article
• Data blending can only be a LEFT join, which means the primary table should contain all possible values.
• Data blending can be more flexible for quickly changing the linking properties on the fly for ad-hoc analysis. For example, data blending allows data sources to be linked on aliases.

Additional Information