Tafsir -

Each item on a survey is treated separately (unless, I suppose, perhaps a qualitative psychological survey of some sort with questions that are used in combinations, I would imagine).  The sample size needed for some will likely be different than others.  For quantitative analyses estimating totals or means or proportions, there are standard deviations and accuracy needs that vary by survey item/question/dependent variable/variable of interest, whatever you call it.  Each such item needs to be considered.  In a survey sampling textbook, you will usually see methods for a single such item.  When doing a survey with multiple items/questions/attributes, you have to make sure that your most vital questions/items are well considered. 

So if I understand your question, then no it does not matter what percent questions are for what data.  If some of these are qualitative questions, though that was not my area, I still cannot see how that could likely matter to your question.   However, for such a mixed methods arrangement, I would think that your individual application needs, your goals, and subject matter should dictate what mix is most reasonable to obtain what you need for your study.  So I doubt that you should try for some mix goal that someone has as a "rule of thumb," if it is not good for your study. 

If your question is, Can you use a different methodology such as stratified random sampling for some questions, and simple random for others? ... that would still depend upon the individual questions/items.   Generally, however, if you can stratify, you will be more accurate in your results.  -  If some questions are best handled with cluster sampling, or any kind of multiple-stage sampling, then this may not be practical to use unless those questions are on a separate survey because of data collection issues.  A cluster sample, whether single- or multi-stage is often done for reasons of data collection logistics. So I doubt you would be referring to having some items/questions on a survey using that methodology combined with anything else.  In probability-based sampling, the sample collection design must be consistent with the estimation methodology, though with model-based methods there may be more flexibility.

As I wrote this, I was realizing all the different things you might have meant by your question, and hope that at least some of the above might be relevant. 

Cheers - Jim