The behavior is UNDEFINED, so if you're going to fill the whole screen anyway, you may consider dropping the clear.. except, on some environments (certain tiling architectures at least), that will actually degrade performance. The full screen clear at the start of rendering is so common operation that I'd be surprised if it wasn't optimized well on all target platforms.
Reason why it's UNDEFINED is that for many architectures, making the content state defined would be an extra overhead, regardless of how you'd define the standard.
As to what you'll find there, I can take a guess based on experience;
On double (or multi-) buffered architectures, like most desktop PC video hardware, you'll likely find the "other" buffer data. This is not guaranteed though, as it's not in the spec, if some weird optimization benefits from it, they will garble the data.
On tiled architectures, you might find the same data as the last frame, some oddly garbled data based on the tile size, or just about anything else.
Then you have some oddball architectures (like the NDS) which might give you just about anything, as their definition of the buffer isn't exactly what you'd expect.