I ... would like to know if there is a preferred way to optimize them.

I don't think one way would be better than then other. It's entirely up to you how to do it. It may also depend on how many filters (or filter actions) you're planning to use.

Also, I seem to recall seeing in the filter help pages that it is preferable to have local storage for filters as opposed to having the mail server sort them out. Why is that?

The filtering happens locally in Thunderbird, not on the server. Thus e.g. filtering on the message body would only work when the entire message has been downloaded, not just the headers. I don't think Thunderbird can do server side filtering.

wrt Organization of filters:

 [] Your considerations should be based upon the particular sender of the email;
     In short... it depends upon a non-standards based naming of email.sending.addresses.
     Since it's non-standard, one can predict chaos... and not really easily filterable content
     which more general filters could address....
    
 [] cf.,  
      [] Some vendors use 1 email address, for all communications:   
                        adverts, accounts, alerts, sign-in, flilght changes, sales...
         In this case, a universal generality (1 folder for HEALTH), are not very helpful, since it is essentially,
         like  a MSoft DOS filesystem model,  is, over time, a large unstructured mix of types of content.
                    I'd classify those vendors as VERY-Unstructured-Email-Sender

        Still, some coarse structuring like HEALTH is still useful, given many other  coarse grain categories;   
        Just be aware that you can get alot of less desirable stuff, yet still from the same vendor.

      HOWEVER,
         [] Not all vendors use one 'sending address' for all email.   
             Some take it to the opposite extreme, having too many email-sending-addresses.

         [] In some cases, a given vendor will have 3rd parties, send emails on their behalf, often survery, feedback...

     [] Many vendors use a mixed send model !    
        Particularly problematic are vendors like airlines, health care, credit reporting agencies, bigBox, Selling-sites...
        In many cases, they have not logically partitioned sending email addresses, by function;

        In some cases , they purposefully, want to make it difficult:
               [] to filter-out the adverts --- that way the can annoyingly be in your face...
               [] instead of what you probably want ...from more critical things like flight changes..., pw related warnings...

Given the above, it is much easier to deal with each vendor alone.

     Thus your filter's could be set up to use Logical folderNames, of your choice(s) 
      (eg., where the structure is your logical partitioning, to try to enforce w rules YOU write. 
      (        this allows YOU to logically group email.   Below the "/" symbol might be YourFolder name.
      (
      (   Note some email progs: Google, Thunderbird...  have differing rules for naming of such "lables".)
      (       Some progs place limits on lable, lengths, characters which can be used. ; or # rules allowed.
      (       Most will be different enough that rules to try to structure "folder-lables" are harder than need be.
      ( 
      ( this could be remedied by actual standards for categories of email address names
                 (      ... don't expect that any time soon ! )

        /PP_who/HEALTH/vendorA/solicitations
        /PP_who/HEALTH/vendorA/Flight or Acct changes....
                                            .....
        /PP_who/HEALTH/vendorB (similar, or... filters that deal with that one particular vendors model.

        /MyBiz_which/Category/subCategory/......
        
it takes a lot of work... but can add some sanity to filters.....  (until vendor changes their email model/subVendor....)

Also consider if filtered mesages need to arrive in inbox, or skip inbox,

    and put into some other filter caregory, of lesser importance. 

OR, if u r like me, and get a ton of email:

     have Logical email "lables-folders"  in a priority-of-importance,  filtered-categories, 
          having those of Ur personal assigned higher priorit(ies) appearing higher
          in the(  emailToolSpecificRules), naming-heirarchy, like:
        /____Critical-Info/ { AcctSecurity.{DataBreach | logins } | 
                                        Travel {Curr | Pending } |
                                        Emerg {onDeath | Drs_who | AutoIns | PP Contacts} |
                                        Finance { BankAlert[i] .....} |
                                       ...
      /__MyReadingList {....}               // note gmail allows each email to be classified by 1..M lables
                                                          //  that allows one to drag a lable onto an existing filtered email
                                                          // allowing a finer grain accessClassification, like this example
                                                          //  but also classification-lables like:  
                                                          //            Alerts....  ToDo.Critical... Fin, nonFin,   
                                                          //            Security, nonSecurity
                                                          // 
       /____A_Status/{ _z V-LargMsgs/{ __GT_4MB | GT 10MB } |  Alerts/{... perhaps the directly above subCategories}
     and those like solicitation-categories occurring  like:
        /zz_Offers/{Retail | Travel | Charity | BankOffers | MostlyUnwanted |  ... }

/------------------

As to "efficiency":

  Most modern ( <10yr old systems), can quite easily handle 1-2K rules, with little lag.
  I think the efficiency of the ability of the end user to easily see/ classify the content is the type of
  efficiency that you should consider as Most important...

My views.... 2Cents