Wide and large Dimension or Survey Factless Fact Table

Hi

I would like some advice on this particular point. We receive responses to surveys containing hundreds of questions. Those surveys are submitted many times to about ten thousands people we follow. Their answers could change in time. Usually the questions aren't associated to numerical value.
What's more, new questions are asked and some aren't asked anymore.

To represent such datas, i could have a wide and quite huge SCD type 2 dimension or a huge factless fact table like described on page 197 in The Datawarehouse Toolkit.

Users want to cross answers to a particular question with answers to others and then identify precisely peoples corresponding.
They also want to filter data from other fact tables with specific answers.

I think the 2 solutions could be used but the SCD seems to me very hard to maintain and the factless very hard to query.

A composite solution could be to duplicate most used questions as an attribute in the dimension and a fact in survey data. Does such a solution sound acceptable ? have you an idea of the performance problems I could encounter.

Thx for your ideas

Guillaume