One is good old objectstore. ie you stream your java class instances into a table, along with an absolute minimum of extra info ClientID and rich/poor may be. Works very well for document based systems where there isn't a lot of cross relations.
Forget things like adhoc queries and reporting and such.
Another approach is to go completely the other way.
You have a table of properties. e.g wealth, debt etc. and then you tie them togther with ClientId. Again very flexible but querying out to a fixed format row is time consuming.
Whether you do the presentation through views, sps, or your classes, you get both the benefits and the costs of a very tightly coupled design, or an ever growing shed load of mappings.
Third option is to have anonymous field names e.g. Text1 Nuber15 Money5 etc. and a type identifier then you map the dn columns to your class properties. To me this gives you the worst of the above two approaches, but some are enamoured of it.
Fourth is you don't actually do poly morphism as such You have clients, RichClients, PoorClients, a lot of relations. More of an amalgamation approach than a polymorphic one and generally my preferred approach.
Last but far from least an unholy confabulation of the above four ideas, ie where you would end up without constant refactoring. :( :( :(
HtHs