Because the PK is the clustered index, inserts are fastest in primary key order. But how bad is the penalty of inserting in random order? How much data needs to be moved in the worst case, the entire table?
Because in an MVCC environment all updates can be considered as inserts, this really keeps me puzzled.
Another question, which is highly related, is how tables are stored in the tablespace. If I first create table1, add some rows, then create table2, add some rows, and then again add some rows to table1 but with smaller PK-value than the rows I inserted before. Then do the rows I inserted before in table1 have to be copied to the end of the tablespace after table2, or does table2 get moved?
There is very little information available about this in the literature, and maybe you would like to clear things up (in a blogpost?).