> how hard is it to guess the 128-bit IDEA key?

Pretty danged hard.

> (I am wondering that in XXX years, computation power would be such 
> that brute force attacks on the random IDEA keys would be a thought.)

Here's a brute-force solution:

 1.  Develop a CPU that can test one billion IDEA keys per second.
 2.  Build a parallel machine using one million of these CPU's.
 3.  Mass produce them cheaply enough that every man, woman and child
     on earth can afford an average of one hundred of the machines.
 4.  Network them all together, a la RSA-129, and start working through
     the 128-bit keyspace.

Assuming ideal performance and no downtime, you should be able to 
exhaustively search the keyspace in a little over 20,000 years.  But
don't let this discourage you--each key will take, on average, only
half that time to break.

--
+------------------+ "Men born to freedom are naturally alert to repel invasion
| James A Campbell |  of their liberty by evil-minded rulers. The greatest
|  Math Sciences,  |  dangers to liberty lurk in insidious encroachment by men
|   U of Memphis   |  of zeal, well-meaning but without understanding."
+------------------+         -- U.S. Supreme Court Justice Louis Brandeis, 1928