Not sure if we are talking about exactly the same thing, but one of the points of using such blocks is a really fast random access to bunch of records located in some bzip-ed chunk.
This works really well if you fulfill some prerequisites (simplified): 1. the pre-compressed, usually tab/space delimited data must be sorted (typical: by chromosome and position) 2. you create an index telling you that the data for chromosome such_and_such in an interval say 2_000_000-2_100_000 is in a chunk 1234 and it this chunk is at the position 987654321 in the file.
As long as all you do is "give me all records and all columns from interval X", this works really fast. On the other hand it can be abused/pushed to imho absurd levels where data files with thousands of columns are queried in that way. Once you got giant say in tens/hundreds of Gb block-compressed files adding new data (that would be extra columns...) to such monstrosities becomes a real PITA.