Yes, one can simulate a control data set by sampling proportional to the census data; provided one has an adequately large population from which to sample. That is some bins (classes) may be oversampled if the population is small. Many convenience data sets tend to be biased (under represented) on the lower and upper age segments. One might consider bootstrapping to get more consistent estimates. Also, one can use the function sample() in R to great effect, rather than runif. With some thought one can make the bootstrap sampling efficient so that one can do tens of thousands of iterations in relatively little time. 

Yes, if you originally recruited adequately large sample size 

Make sure that stratification does not affect your sample 

What is your research question?

RG deleted the frist part of the question. I have 5 group of patients with different diseases, the age in these groups is not normally distribuited and varies. I tested the difference of distribution and median location between sexes with k-s and wilcoxon. I would to test the H0 that each group is sampled without differences from the general population with the same tests. But I do not have normal controls. .... How can I use the census data?

Thanks to  mr. Tromp, but i still miss something,maybe you could add something after reading the frist part of the q. I found a couple of papers i'm going to read an then  share with you

I agree with Gerard. Bootstrap seems to be a good option. sample() in R is also a nice function. If I remember it correctly you can also use this function for stratified sampling e.g. to ensure a similar sex-ratio in your controls compared to your disease groups.

Ok , I can sample my total data proportional to census data and create a control group to test my hypothesys in each subgroup with the due corections. Or I can  also bootstrap a W statistic. I will try both and see.