You cannot separate the BLOB from the work table and this can not be implemented for any PRPC Application.
When ever a work object is saved in DB the following events occur in a sequence.
All the work object information is first stored in the blob.
Exposed table columns that require an update are identified.
Identified columns are populated with the latest data found in the blob.
Now its obvious why you cannot store the blob separately. As far as data is concerned everything revolves around the blob in PRPC.
To address the problem you have; try the following ...
Know the data retention policy of your client and purge any work object data that is older than expected number of months.
Place a partition in your DB tables for every 3 months. Discuss with your DBA on how this will help your application.
The size of the pyWorkPage should not exceed 5 MB; you could focus on staying within this limit. ( Usually a pyWorkPage will not grow beyond 100 KB. However in some applications the request and response XMLs of SOAP interactions are stored within the work object is such cases the pyWorkPage size could get out of control.)