Question: Non/linear fit of near zero data points

I have a data point set:

relativní_tlak:=<0.063018,0.078419,0.119628,0.159668,0.199865>:

ads_mnozstvi:=<0.001467443666,0.001528693055,0.001659939952,0.001774105924,0.001883866808>:

I wanna make the least square fit with function in form of:

bet:=(x/(1-x))/((1/(a*b))+(((b-1)/(a*b))*x)):

also the command looks like:

with(Statistics):bet_nejmensi_ctverce:=NonlinearFit(bet,relativní_tlak,ads_mnozstvi,x,parameternames=[a,b],output=parametervector);

I found out the least square fit is not correct using the near-zero values in 'ads_mnozstvi' vector (the parameters have no physical sense).

The correct a,b values can be obtained when 'ads_mnozstvi' values are multiplied by 1000. The correct a,b values thus are:

[1.56257913677048,108.232791649022]

Why Maple cannot make correct fit with near-zero values?