SQL Server has frankly always had some internal issues with parallelizing queries. Parallelization is great for (re)building indexes, but for queries it's always had its quirks.

Unless you are absolutely sure the tables are properly indexed, use MAXDOP to drop the parallelization to a small number, 2-3, to help keep SQL from getting "lost". Of course for very large tables you'll need to bump it up some from that, but I still wouldn't let MAXDOP = 0 be specified or implied.