The 35 l/h was just a fit for the calculations. Depending on the P/Q curve you should expect a 70% - 50% flow drop.
I imagine that you have already read the related articles here. You should also take a look at BillA's work over at
overclockers.com.
On the subject of measuring flow. You should get a decent estimate (let's say a 10% error) if you make relative measurments. First measure the flow at zero head pressure. Now you have a reference point for which we also know the real flow (rated value of the pump). All subsequent measurments (at different head pressures, meaning also different flow rates) can than be express in relevance to this value.
The only problem is that in our case the flow resistance increases by the square of fluid velocity. This hols true if Reynold's number is > 1000 (between 1 in 1000 it's neither linear nor square).
Let flow be 500 l/h and hose diameter 1/2" = 12cm
v = (dV/dt) / S
v - avarage water velocitiy
S - hose cross-section
dV/dt - volume flow
Re = [ro] * v * d / [ni] = [ro] * (dV/dt) / S * d / [ni]
Re - Reynold's number
[ro] - desity of water (1000kg/m^3)
d - fluid layer thickness (equal to the hose diameter)
[ni] - viscosity of water (10^-3 Ns/m^2)
Re = (1000 kg/m^3) * (500/3600/1000 m^3/s) / ([pi] * (0.06m)^2) * 0.12 / 10^-3 = 1473
Unfortunatly the square aproximation (even worse with low flow rates becouse Re can in that case drop well below 1000) will in the end account for most of the error.