You are correct in that diaper memory, oh, I mean hyper memory, is a very low budget way to advertise you have a decent card to get people to pay more, then go ahead and cripple it so the performance isn't there.
When you have onboard video memory it is usually GDDR2 or better yet GDDR3. This is very fast memory, that you already noted is on the GPU board so 1. it is faster. 2. It is a very short distance from the GPU itself. 3. It is directly controlled by the GPU. 4. It is dedicated to the GPU.
Hyper Memory is shared system memory, that means it is the RAM your system has in it. Coupe drawbacks to this are:
1. It is system memory, and is being used and controlled by the system. It may not be available to allocate when you need it, or you may not have enough system meory to start with.
2. It is slower than the DDR2 counterpart (especially if you are talking about GDDR3 (the G-stands for graphics))
3. In an Intel system the off board memory controller puts 2 more hops in your GPU getting to the memory if it is even available. From GPU to CPU to Memory Controller to Memory latency.
4. Distance from GPU to CPU to Memory Controller to RAM, and all the way back. Remember, this stuff is doing millions of calcs a second and what seems like a short distance becomes a very long round trip when you are processing that many instructions a sec.
It is overall much slower and highly NOT recommended, but far better than an onboard integrated GPU. If you got a good deal I would just say, "Hey, it's a little bit better than nothing." If you specifically bought a system for doing some moderate gaming, and thought you were getting something else, and paid a little more for it, return the thing and get a differnt system.