I believe what's happening here is very low Reynold's number laminar flow, which means that there are many parallel layers of "sticky" (or viscous) fluid. As the handle is turned, the dyes remain within their original layers and do not mix with each other, even as they are spread out over their individual layers. As the crank is turned backwards, the process is inverted, and since there is no fluid turbulence in laminar flow, the process can be inverted almost perfectly. - See more at: http://www.stevespanglerscience.com/lab/experiments/twist-in-time-laminar-flow#sthash.w6XWDbhH.dpuf