Private Shared Variables Vs Local / Namespace Performance

I came across a number of new Private Shared variables (of type Hashtables(Of String), initialized in the declaration) added to a partial class for a very large (DataContext-derived) class. This seems sensible to me in one sense because they never change, and making these shared variables ensures that they won't get re-initialized every time a function is called. However, these variables are only used within the scope of one function in the class, and I fear the private namespace of this DataContext-derived class is getting rather polluted, and having these sorts of things exposed at such a high level might be confusing to others reading the code in the future.

Would there be negative performance impact to making these local variables within the function where they are used, or is there some better way to handle this? Basically we are using these 3 hashtables to determine whether anything within particular subsets of properties changed (using GetModifiedMembers and then using the Overlaps function of the hashset to see if any of the modified members correspond to members we care about).

Edit: I caved and took the time to write my own test program, which confirmed that there is a cost to using local variables (which I assume applies generally to all cases -- I doubt there's any case where a shared variable would be slower unless using the shared variable requires some additional logic to do so properly): [Code]

So in this particular case, using the local variable costs about 200%. But in most cases (including my own), the time is probably negligible compared to the overall task. So I guess the question now becomes, how do people generally feel about improving code maintainability at the cost of negligible but known performance impacts?