Probability theory excercise: Buy 20 components for a project. At 30% fault rate for each component, what is the probability that at least one in the project is faulty? If you spend, on average, 10 hours of debugging caused by each faulty component, how many hours (expected value) are you going to spend on this 20-component project?
My experience on my own projects as well as debugging a lot of others' projects is that while it often works to buy an Arduino board, an LED and a resistor from Ebay to blink the LED, when the project gets any more complex than that, the chances that at least one vital component is broken in a funny way asymptotically reaches 100%. For example, it takes some debugging to find out why a 3V3 regulator is giving out 1V8. As a beginner, you always doubt your own work, which is correct because, when using genuine components, you are the source of 99.9% of the problems. But when you mix in a bagfull of faulty components you need to debug and swap, working for hours and hours on them just to find that they were faulty to begin with, then order some more, wait for a week to get the new parts - which again may be broken - you are creating an environment where you'll never get your project finished until you lose all your motivation.
Another issue is that you typically cannot source all of the components needed in a project from Ebay, so you "need" to buy from a proper distributor anyway.
Ebay is great for sourcing some trivial parts that for some weird reason are missing from the distributors, are hard to find, or are prohibitively expensive for what they are. For example, I have bought several $1 flux pens when FARNELL was, at the time, only selling flux pens for $100/ea(!).