I've done this sort of thing before with traffic generators and measuring the transfer between PCs. The big thing to remeber is that every transport mechanism introduces an overhed. What you need to do is try out various traffic generation methods, transfer methods and software/PC combinations to get the maximum throughput you can achieve on a test network (say both devices back-to-back in a good switch on the bench). That's the maximum throughput you can achieve/measure. Then when you repeat the test in a real environment if you get lower throughput you can give a definitive answer knowing your test method isn't the limiting factor. If you get the same, you'll still be able to say "the bandwidth is at least x".