Please keep the email in the loop.
flexible model is much more expensive than rigid one. If you are not interested in intra-molecular vibration, which is probably the case for shear-viscosity, please use a rigid water model instead.
If you are using a flexible model, time step at 1 fs is too long to resolve the vibration. It should be at most 0.25 fs.
1 fs is ok for rigid model.
The paper I referred studies rigid models, so their recommendation to sampling time/correlation length/simulation time applies to rigid ones only. I can not guarantee the same setting works in your system.
To give you a head start, try the following settings:
variable dt equal 0.001
variable p equal 5000
variable s equal 1
variable d equal $p*$s