The EXECUTION_CONTEXT Blob Column in LSW_TASK_EXECUTION_CONTEXT holds the runtime data of an Task, it seems to be a compressed java serialized object, it may be possible to un-compress it with oracle functions and then read it using tools like jdserialize, but don't think the data will be usable, Direct Process Server DB queries are not recommended since IBM may change the structure and working of these tables and the data within in future versions. Also they may cause locks hindering the performance of Process Server.
You can use REST API's, or JS API based retrievals to get that data
The following oracle query will return a list of context sizes for tasks
select distinct LT.ACTIVITY_NAME as ActivityName, LS."NAME" as SnapshotName, LP.SHORT_NAME as ProjectAcronym, LBI.BPD_REF as BPDReferenceID, LBI.TIP as isTip,
LBI.BPD_NAME as BPDName, LP.NAME as applicationName, count(DISTINCT LBI.BPD_INSTANCE_ID) as instanceCount, max(dbms_lob.getlength(LTEC.EXECUTION_CONTEXT)) as contextSize
from lsw_bpd_instance lbi, LSW_PROJECT lp, LSW_SNAPSHOT ls, LSW_BPD_INSTANCE_DATA LBID, LSW_TASK LT, LSW_TASK_EXECUTION_CONTEXT LTEC
where LBI.SNAPSHOT_ID = LS.SNAPSHOT_ID
and LS.PROJECT_ID = LP.PROJECT_ID
and LBI.BPD_INSTANCE_ID = LBID.BPD_INSTANCE_ID
and dbms_lob.getlength(LTEC.EXECUTION_CONTEXT) > ?
and LT.BPD_INSTANCE_ID = LBI.BPD_INSTANCE_ID
and LT.TASK_ID = LTEC.TASK_ID
group by LT.ACTIVITY_NAME, LS."NAME", LP.SHORT_NAME, LBI.BPD_REF, LBI.TIP,
LBI.BPD_NAME, LP.NAME
order by count(DISTINCT LBI.BPD_INSTANCE_ID) DESC