I think what you're asking is:
When rendering your frame, you render some data to a texture. Then you want to use this texture data in subsequent runs of a vertex shader (the same three texture samples for all runs of the vertex shader). Normally this would make sense to be a shader constant; but in your case it can't be because the data was generated on the GPU earlier in that frame and is never pulled back to the CPU. So you're basically searching for a way to use this data in your vertex shader, without having every vertex shader run have to take the 3 samples.
Is that correct?
I don't know of a way to do that, but:
1) Could you instead calculate this data on the CPU? Or do the calculations that go into those 3 pixels of interest in your texture come from something that has to be done on the GPU? Or,
2) Could you put up with a delay of a couple of frames, allowing you to pull this data back to the CPU when it's available and then set it as a shader constant?
Also, is doing the 3 samples in every VS run an actual a performance problem, or an imagined one?