I'm having a similar problem.  Much smaller tables, about 6 columns, 140000 rows.  Looking at the query plan, the read from the inserted table is by far the most expensive operation, even though there is only a single row in it.  Other operations that are kicked off by the trigger, which should be much more expensive (READTEXT / WRITETEXT along with finding the block of text to replace) take a significant amount less time.  This makes absolutely no sense.  I've even tried to eliminate any joins at all (setting the values into params and then using the params to find the data that I need in other tables.  The results are still pathetic.  I am very curious if you ever found a solution for this, or maybe after 3 months since the last post, somebody has come across some new information about what is going on with the inserted table being so inefficient to read.