What's to stop someone from simulating that same sensory input to the roach brain without the robot body? How would the roach know the difference?
The roach brain would be a physical brain. Not a simulated brain. There is a huge difference. I'm NOT talking about the Matrix where we're all brains floating around in a vat hooked up to a bunch of electrodes. Rather, the simulation hypothesis proposes that our brains would be part of the simulation. Not a physical computer architecture that's setup mimic a brain - but instead a full on simulation.
We can simulate lots of things. But none of those simulations have the properties the actual things have. In the near future we're going to simulate a human brain (The Human Brain Project). But that doesn't mean the simulated brain will have the properties of actual brains (e.g. subjective first person experience). It doesn't mean it will be able to experience pain, or feel pleasure, or love. It will be a simulation of how a brains work. Not an actual working brain.
Simulating an earthquake inside a supercomputer doesn't cause actual tremors inside the server room. Simulating a blizzard doesn't produce actual snow. And I have yet to hear a good reason why simulating a brain would produce actual consciousness.
Perhaps we could build a physical brain that would be conscious. But that's not the same thing as a simulation of a brain that is conscious.