I am in the situation to test several "select *" kind of SPARQL queries
against few TB of triple data, of course not in production.
However, we only have limited machine resources (4GB memory) to test the queries.
I understand that it requires more memory, but are there any alternatives to run queries and get results? (Time consuming is welcome)
My laptop has 32GB of RAM, so that sounds under-resourced even for a dev server. Having said that, for any particular query I would look for ways to reduce the number of triples you're running against. Are your triples segmented into graphs, and if so, can a query be directed against one graph? Another reducing strategy is to use the $query parameter to sem:sparql to identify documents that hold the triples you care about.