This is a great question but a bit complex to answer in this type of forum. The quick answer: a full table scan is sometimes a direct path read. It depends on your database version (adaptive direct path reads appeared in 11g), table size, parallelism, and other factors. If you're interested in the topic. Frits Hoogland has done quite a bit of work on this topic, and has an excellent presentation on it: http://fritshoogland.files.wordpress.com/2012/06/about-multiblock-reads-v2.pdf
A direct path read bypasses the buffer cache --- data is read from the database directly to the server process'es PGA. A Full Table Scan may be a direct path read if it is not small. Reads using Parallel Query are direct path reads.