Start on the target, and find the spid for the SSIS component running. Try to determine most common wait type. If it's PageLatchIO or another variant on disk waits, you're fine there. If it's something else you need to inspect that. Also, the target either is in simple mode or has the log file grown out to the size you need, right? You don't want to be auto-growthing in the middle of something that large.

My usual technique for that is simply to ping sys.sysprocesses once/second and dump the results to a table for 2 or 3 minutes. Take a look at the results of that.

Why are you loading two sources at once, and are you merging the streams prior to insertion? If you're paralleling the process then you're contending.

Otherwise I'd really have to dig into details. What's the SAN target, width of the stream, any implicit conversions going on during delivery, etc etc. Something that large I always do dev testing with smaller result sets, say, 2 million rows, from the source (filter it). At least that gives me a generic baseline expectation.